{ "$schema": "https://www.yerel.ai/schemas/benchmarks-v1.json", "version": "2025.05", "generatedAt": "2026-11-01T19:08:03.881Z", "license": "CC-BY-4.0", "source": "https://www.yerel.ai/api", "methodology": "https://www.yerel.ai/benchmark/metodoloji", "qualityGate": { "score": 9.9, "target": 9.9, "passed": true, "summary": "9.9/10 öz-rubrik puanı (kendi rubrikimiz, bağımsız denetim henüz yok) · 47 model, 9 donanım profili, 2 birinci-elden ölçülen kalite boyutu (+ TPS); diğer skorlar atıflı/planlı", "checks": [ { "id": "schema", "label": "Typed run schema", "score": 2, "max": 2, "passed": true, "detail": "91 runs checked, 0 blocking error(s)" }, { "id": "statistics", "label": "Reproducible (deterministic n=1 + SHA-256)", "score": 1.5, "max": 1.5, "passed": true, "detail": "64/64 birinci-elden koşu yeniden üretilebilir: kalite boyutları logprob + sabit tohum (seed 42) ile deterministik; TPS koşuları duvar-saati ölçümüdür (sabit komut + SHA-256, tohum-deterministik değil)" }, { "id": "methodology", "label": "Methodology is documented", "score": 1.5, "max": 1.5, "passed": true, "detail": "14 dimensions define samples, scoring, and range metadata" }, { "id": "reproducibility", "label": "Reproducibility card coverage", "score": 1.5, "max": 1.5, "passed": true, "detail": "14/14 dimensions have reproducibility audits" }, { "id": "evidence", "label": "Evidence state is disclosed", "score": 1.4, "max": 1.5, "passed": true, "detail": "23 runs are explicitly provisional until raw logs and output hashes are published" }, { "id": "public-export", "label": "Public export contract", "score": 1, "max": 1, "passed": true, "detail": "JSON, CSV, dimensions, hardware, and merged community feeds are generated at build time" }, { "id": "community-gate", "label": "Community quarantine gate", "score": 1, "max": 1, "passed": true, "detail": "Community submissions are validated, deduped, and invalid rows are quarantined before merge" } ] }, "counts": { "total": 91, "editorPicks": 68, "verified": 0, "community": 20, "pending": 0, "modeled": 4, "measured": 64, "modelsCovered": 47, "hardwareCovered": 9, "dimensionsCovered": 5, "totalRepetitions": 146, "measuredQuality": 38, "measuredSpeed": 26, "measuredQualityDimensions": 2, "citedQualityDimensions": 1, "dimensionsWithAnyRun": 5 }, "runs": [ { "id": "trendyol-asure-12b-cost-001", "modelSlug": "trendyol-llm-asure-12b", "dimension": "cost-per-1m-tr", "scores": [ 3.37 ], "sampleCount": 1, "hardware": "rtx-5080-16gb", "runtime": "ollama-0.24", "quantization": "mixed_q4_q6_k", "contextLength": 2048, "seed": 0, "datasetVersion": "cost-2025.05", "submittedAt": "2026-10-02T00:00:00Z", "submittedBy": "yerelai-editor", "status": "editor-pick", "verifications": 1, "measurement": "modeled", "notes": "Modellenmiş tahmin, ölçüm değil. Elektrik-only alt sınır: 1M token / 79.0 tok/s (RTX 5080 16GB üzerinde ölçülen hız, koşu trendyol-asure-12b-tps-rtx5080-mixed-20260528) = 3.52 saat; 0.360 kW (GPU TDP varsayımı, gerçek güç çekişi ölçülmedi) × ₺2.66/kWh (varsayım) × 3.52 saat ≈ ₺3.37. Donanım satın alma dahil değildir; ROI için canlı tam sistem sepeti gerekir." }, { "id": "gpt-4o-cost-api-001", "modelSlug": "gpt-4o", "dimension": "cost-per-1m-tr", "scores": [ 365 ], "sampleCount": 1, "hardware": "api", "runtime": "api-direct", "quantization": "api", "contextLength": 8192, "seed": 0, "datasetVersion": "cost-2025.05", "submittedAt": "2026-05-20T09:01:00Z", "submittedBy": "yerelai-editor", "status": "editor-pick", "verifications": 1, "measurement": "modeled", "notes": "Modellenmiş tahmin, ölçüm değil. $2.50/M input + $10/M output × 33.6 ₺/USD × 1.4 TR-token overhead, 30/70 input/output blend." }, { "id": "claude-3-5-sonnet-cost-001", "modelSlug": "claude-3-5-sonnet", "dimension": "cost-per-1m-tr", "scores": [ 196 ], "sampleCount": 1, "hardware": "api", "runtime": "api-direct", "quantization": "api", "contextLength": 8192, "seed": 0, "datasetVersion": "cost-2025.05", "submittedAt": "2026-05-20T10:31:00Z", "submittedBy": "yerelai-editor", "status": "editor-pick", "verifications": 1, "measurement": "modeled", "notes": "Modellenmiş tahmin, ölçüm değil. $3/M in + $15/M out × 33.6 × 1.4 × (30/70 blend)." }, { "id": "llama-3-3-70b-cost-api-001", "modelSlug": "llama-3-3-70b", "dimension": "cost-per-1m-tr", "scores": [ 42 ], "sampleCount": 1, "hardware": "api", "runtime": "api-direct", "quantization": "fp8", "contextLength": 8192, "seed": 0, "datasetVersion": "cost-2025.05", "submittedAt": "2026-05-22T10:31:00Z", "submittedBy": "yerelai-editor", "status": "editor-pick", "verifications": 1, "measurement": "modeled", "notes": "Modellenmiş tahmin, ölçüm değil. Groq fiyatlandırması $0.59/M in + $0.79/M out." }, { "id": "kumru-2b-tps-rtx5080-q4-20260528", "modelSlug": "kumru-2b", "dimension": "tps", "scores": [ 446, 397.42, 443.66, 406.67, 454.33 ], "sampleCount": 5, "hardware": "rtx-5080-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 2048, "seed": 42, "datasetVersion": "tps-ollama-2026.05", "submittedAt": "2026-05-28T05:13:10.545Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/local-ollama-tps-turkish-sweep-20260528-v3.json", "outputHash": "897d20358d0a41bb640d93209207220980320a91618c1f3a18a382dbbe22e414", "notes": "Ollama tag alibayram/kumru:latest, digest 04f3c85761dd, params=2.4B. Batch=1, num_ctx=2048, num_predict=512, temperature=0, seed=42, 5 reps on NVIDIA RTX 5080 16GB driver 595.97. First run stopped at 448 generated tokens; remaining runs hit length." }, { "id": "turkcell-llm-7b-tps-rtx5080-q4-20260528", "modelSlug": "turkcell-llm-7b", "dimension": "tps", "scores": [ 146.35, 145.77, 144.04, 145.08, 144.99 ], "sampleCount": 5, "hardware": "rtx-5080-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 2048, "seed": 42, "datasetVersion": "tps-ollama-2026.05", "submittedAt": "2026-05-28T05:13:10.545Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/local-ollama-tps-turkish-sweep-20260528-v3.json", "outputHash": "897d20358d0a41bb640d93209207220980320a91618c1f3a18a382dbbe22e414", "notes": "Ollama tag RefinedNeuro/Turkcell-LLM-7b-v1:latest, digest 50f61cb255fa, params=7.4B. Batch=1, num_ctx=2048, num_predict=512, temperature=0, seed=42, 5 reps on NVIDIA RTX 5080 16GB driver 595.97. Model stopped naturally at 494/500 tokens." }, { "id": "ytu-turkish-gemma-9b-tps-rtx5080-q4-20260528", "modelSlug": "ytu-turkish-gemma-9b", "dimension": "tps", "scores": [ 100.61, 100.79, 101.65, 101.14, 101.3 ], "sampleCount": 5, "hardware": "rtx-5080-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 2048, "seed": 42, "datasetVersion": "tps-ollama-2026.05", "submittedAt": "2026-05-28T05:13:10.545Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/local-ollama-tps-turkish-sweep-20260528-v3.json", "outputHash": "897d20358d0a41bb640d93209207220980320a91618c1f3a18a382dbbe22e414", "notes": "Ollama tag alibayram/turkish-gemma-9b-v0.1:latest, digest c51ed67b59a4, params=9.2B. Ollama metadata source is ytu-ce-cosmos/Turkish-Gemma-9b-v0.1. Batch=1, num_ctx=2048, num_predict=512, temperature=0, seed=42, 5 reps on NVIDIA RTX 5080 16GB driver 595.97." }, { "id": "trendyol-asure-12b-tps-rtx5080-mixed-20260528", "modelSlug": "trendyol-llm-asure-12b", "dimension": "tps", "scores": [ 78.38, 79.06, 78.59, 79.61, 79.47 ], "sampleCount": 5, "hardware": "rtx-5080-16gb", "runtime": "ollama-0.24", "quantization": "mixed_q4_q6_k", "contextLength": 2048, "seed": 42, "datasetVersion": "tps-ollama-2026.05", "submittedAt": "2026-05-28T05:13:10.545Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/local-ollama-tps-turkish-sweep-20260528-v3.json", "outputHash": "897d20358d0a41bb640d93209207220980320a91618c1f3a18a382dbbe22e414", "notes": "Ollama tag alibayram/Trendyol-LLM-Asure-12B:latest, digest 592bfbc20577, params=11.8B. Ollama summary reports quantization=unknown; verbose tensor metadata shows a mixed Q4_K/Q6_K GGUF, so this row is labeled mixed_q4_q6_k rather than pretending it is Q4_K_M. Batch=1, num_ctx=2048, num_predict=512, temperature=0, seed=42, 5 reps on NVIDIA RTX 5080 16GB driver 595.97." }, { "id": "rn-tr-r1-tps-rtx5080-q4-20260528", "modelSlug": "rn-tr-r1", "dimension": "tps", "scores": [ 133.66, 133.22, 133.07, 134.07, 133.63 ], "sampleCount": 5, "hardware": "rtx-5080-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 2048, "seed": 42, "datasetVersion": "tps-ollama-2026.05", "submittedAt": "2026-05-28T05:13:10.545Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/local-ollama-tps-turkish-sweep-20260528-v3.json", "outputHash": "897d20358d0a41bb640d93209207220980320a91618c1f3a18a382dbbe22e414", "notes": "Ollama tag RefinedNeuro/RN_TR_R1:latest, digest f20cdcf454d7, params=8.0B. Batch=1, num_ctx=2048, num_predict=512, temperature=0, seed=42, 5 reps on NVIDIA RTX 5080 16GB driver 595.97." }, { "id": "rn-tr-r2-tps-rtx5080-q4-20260528", "modelSlug": "rn-tr-r2", "dimension": "tps", "scores": [ 133.44, 133.15, 132.65, 133.58, 133.51 ], "sampleCount": 5, "hardware": "rtx-5080-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 2048, "seed": 42, "datasetVersion": "tps-ollama-2026.05", "submittedAt": "2026-05-28T05:13:10.545Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/local-ollama-tps-turkish-sweep-20260528-v3.json", "outputHash": "897d20358d0a41bb640d93209207220980320a91618c1f3a18a382dbbe22e414", "notes": "Ollama tag RefinedNeuro/RN_TR_R2:latest, digest 8eff3a89286d, params=8.0B. Batch=1, num_ctx=2048, num_predict=512, temperature=0, seed=42, 5 reps on NVIDIA RTX 5080 16GB driver 595.97." }, { "id": "mistral-7b-turkish-tps-rtx5080-q5-20260528", "modelSlug": "mistral-7b-turkish", "dimension": "tps", "scores": [ 130.21, 130.36, 130.93, 129.42, 130.37 ], "sampleCount": 5, "hardware": "rtx-5080-16gb", "runtime": "ollama-0.24", "quantization": "q5_k_m", "contextLength": 2048, "seed": 42, "datasetVersion": "tps-ollama-2026.05", "submittedAt": "2026-05-28T05:13:10.545Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/local-ollama-tps-turkish-sweep-20260528-v3.json", "outputHash": "897d20358d0a41bb640d93209207220980320a91618c1f3a18a382dbbe22e414", "notes": "Ollama tag koezgen/malhajar-mistral-tur7b:latest, digest 45c41d844f56, params=7.2B. Batch=1, num_ctx=2048, num_predict=512, temperature=0, seed=42, 5 reps on NVIDIA RTX 5080 16GB driver 595.97." }, { "id": "qwen-2-5-7b-tps-rx7900xtx-q4-20260601", "modelSlug": "qwen-2-5-7b", "dimension": "tps", "scores": [ 116.04, 115.44, 115.27, 115.65 ], "sampleCount": 4, "hardware": "amd-rx7900xtx-24gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T19:53:16Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rx7900xtx-tps/capture.json", "outputHash": "bf5861a1dc0e8460ef5201d736412b0a2483c58fecfbcc9d1ad5d3787e7d46ae", "notes": "RX 7900 XTX (24GB), ROCm 7.10, HIPBLAS, Ollama 0.24.0, Windows 10. 4 reps with warmup discarded, variance 0.2 %. Prefill_tps median 1277 t/s, TTFT median 30.5 ms." }, { "id": "mistral-7b-v0-3-tps-rx7900xtx-q4-20260601", "modelSlug": "mistral-7b-v0-3", "dimension": "tps", "scores": [ 118.79, 118.72, 118.38, 119.28 ], "sampleCount": 4, "hardware": "amd-rx7900xtx-24gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T19:53:16Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rx7900xtx-tps/capture.json", "outputHash": "bf5861a1dc0e8460ef5201d736412b0a2483c58fecfbcc9d1ad5d3787e7d46ae", "notes": "RX 7900 XTX (24GB), ROCm 7.10, HIPBLAS, Ollama 0.24.0, Windows 10. 4 reps with warmup discarded, variance 0.4 %. Prefill_tps median 511 t/s, TTFT median 29.3 ms." }, { "id": "llama-3-1-8b-tps-rx7900xtx-q4-20260601", "modelSlug": "llama-3-1-8b", "dimension": "tps", "scores": [ 109.37, 109.49, 109.25, 109.3 ], "sampleCount": 4, "hardware": "amd-rx7900xtx-24gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T19:53:16Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rx7900xtx-tps/capture.json", "outputHash": "bf5861a1dc0e8460ef5201d736412b0a2483c58fecfbcc9d1ad5d3787e7d46ae", "notes": "RX 7900 XTX (24GB), ROCm 7.10, HIPBLAS, Ollama 0.24.0, Windows 10. 4 reps with warmup discarded, variance 0.1 %. Prefill_tps median 724 t/s, TTFT median 27.6 ms." }, { "id": "rn-tr-r2-tps-rx7900xtx-q4-20260601", "modelSlug": "rn-tr-r2", "dimension": "tps", "scores": [ 109.2, 109.27, 109.09, 108.41 ], "sampleCount": 4, "hardware": "amd-rx7900xtx-24gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T19:53:16Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rx7900xtx-tps/capture.json", "outputHash": "bf5861a1dc0e8460ef5201d736412b0a2483c58fecfbcc9d1ad5d3787e7d46ae", "notes": "RX 7900 XTX (24GB), ROCm 7.10, HIPBLAS, Ollama 0.24.0, Windows 10. Türkçe fine-tune of base Llama. 4 reps with warmup discarded, variance 0.4 %. Prefill_tps median 2898 t/s (very high), TTFT median 30.4 ms." }, { "id": "llama-3-2-11b-vision-tps-rx7900xtx-q4-20260601", "modelSlug": "llama-3-2-11b-vision", "dimension": "tps", "scores": [ 97.94, 98.31, 97.82, 97.5 ], "sampleCount": 4, "hardware": "amd-rx7900xtx-24gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T19:53:16Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rx7900xtx-tps/capture.json", "outputHash": "bf5861a1dc0e8460ef5201d736412b0a2483c58fecfbcc9d1ad5d3787e7d46ae", "notes": "RX 7900 XTX (24GB), ROCm 7.10, HIPBLAS, Ollama 0.24.0, Windows 10. Text-only run; vision tokens not exercised. 4 reps with warmup discarded, variance 0.4 %. Prefill_tps median 601 t/s, TTFT median 31.6 ms." }, { "id": "kumru-2b-tps-rtx3080laptop-q4-20260601", "modelSlug": "kumru-2b", "dimension": "tps", "scores": [ 174.23 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "7effe7c5bf9fb4d20a8f399280cc1c25565c79d25135b15121636fc85fa37585", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi alibayram/kumru:latest; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 173.87, p5 171.65, p95 175.59 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "gemma-3-1b-it-tps-rtx3080laptop-q4-20260601", "modelSlug": "gemma-3-1b-it", "dimension": "tps", "scores": [ 160.42 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "686a16dfa5f9805daf2179c826767958486b0fa299819350b2f09b33cd30f4b0", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi gemma3:1b; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 160.66, p5 159.68, p95 161.96 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "phi-3-5-mini-tps-rtx3080laptop-q4-20260601", "modelSlug": "phi-3-5-mini", "dimension": "tps", "scores": [ 155.44 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "8f494f4aa9db60d8ec59160449e1a872cebad444d6ad4749b220cecfc1d040b0", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi phi3.5:3.8b; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 155.11, p5 152.41, p95 157.35 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "gemma-3-4b-it-tps-rtx3080laptop-q4-20260601", "modelSlug": "gemma-3-4b-it", "dimension": "tps", "scores": [ 97.65 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "093986892faf5744f43190370c951328c1f98edac93358ab5f3bfcb7987ba171", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi gemma3:4b; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 97.68, p5 97.17, p95 98.21 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "mistral-7b-v0-3-tps-rtx3080laptop-q4-20260601", "modelSlug": "mistral-7b-v0-3", "dimension": "tps", "scores": [ 89.64 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "399886c632ff8c6bfabf8c1cfa73757dc6a5155a674404851fb599544e647929", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi mistral:7b; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 89.58, p5 87.89, p95 91.17 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "mistral-7b-turkish-tps-rtx3080laptop-q4-20260601", "modelSlug": "mistral-7b-turkish", "dimension": "tps", "scores": [ 87.27 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "c1e1f2596f203477d7b29c2a2ac0a78accc62cb736acab90a862a143605f6efc", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi koezgen/malhajar-mistral-tur7b:latest; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 86.61, p5 82.68, p95 89.61 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar. Not: aynı Ollama etiketi 2026-05-28 RTX 5080 taramasında Ollama üstverisine göre Q5_K_M olarak kaydedildi." }, { "id": "turkcell-llm-7b-tps-rtx3080laptop-q4-20260601", "modelSlug": "turkcell-llm-7b", "dimension": "tps", "scores": [ 85.77 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "15c735314f30a3bf260693b8763cf094bbe1210afc3088060f2f4c5834f6dad0", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi RefinedNeuro/Turkcell-LLM-7b-v1:latest; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 85.75, p5 85.41, p95 86.06 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "qwen-2-5-7b-tps-rtx3080laptop-q4-20260601", "modelSlug": "qwen-2-5-7b", "dimension": "tps", "scores": [ 80.42 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "0a3b51c3ef429fd8fd597cd7475ab6f28bf7d89549a51ace0a0b91f1327a46f7", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi qwen2.5:7b; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 80.38, p5 79.03, p95 81.68 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "rn-tr-r1-tps-rtx3080laptop-q4-20260601", "modelSlug": "rn-tr-r1", "dimension": "tps", "scores": [ 79.89 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "2fcf366fa036f66750f0331d1579c19b269b75080b2049989268457cd73e0cc0", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi RefinedNeuro/RN_TR_R1:latest; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 80.05, p5 79.49, p95 80.84 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "rn-tr-r2-tps-rtx3080laptop-q4-20260601", "modelSlug": "rn-tr-r2", "dimension": "tps", "scores": [ 79.26 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "2adcda552c27a5715f0886586c27076ae44936b7814d6be068daab06dcc6114c", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi RefinedNeuro/RN_TR_R2:latest; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 79.21, p5 78.85, p95 79.51 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "llama-3-2-11b-vision-tps-rtx3080laptop-q4-20260601", "modelSlug": "llama-3-2-11b-vision", "dimension": "tps", "scores": [ 67.01 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "11ac2c921fad5de0229df8ce26154a4431aa56ed8c5804a22b525cd4b087141d", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi llama3.2-vision:11b; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 67.03, p5 67, p95 67.1 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "ytu-turkish-gemma-9b-tps-rtx3080laptop-q4-20260601", "modelSlug": "ytu-turkish-gemma-9b", "dimension": "tps", "scores": [ 66 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "8ac298362c2b588acc5385742377f19b79107e4559c29da03465a044c5ffbe63", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi alibayram/turkish-gemma-9b-v0.1:latest; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 66.14, p5 65.66, p95 66.81 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "trendyol-llm-asure-12b-tps-rtx3080laptop-q4-20260601", "modelSlug": "trendyol-llm-asure-12b", "dimension": "tps", "scores": [ 43.35 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "618ca08b4886de870773e645b514159a5fd187bb8962846ccaa859ea6d093574", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi alibayram/Trendyol-LLM-Asure-12B:latest; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 43.32, p5 43.14, p95 43.45 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar. Not: aynı Ollama etiketi 2026-05-28 RTX 5080 taramasında tensör üstverisine göre karışık Q4_K/Q6_K GGUF olarak kaydedildi." }, { "id": "gemma-3-12b-it-tps-rtx3080laptop-q4-20260601", "modelSlug": "gemma-3-12b-it", "dimension": "tps", "scores": [ 43.31 ], "sampleCount": 5, "hardware": "rtx-3080-laptop-16gb", "runtime": "ollama-0.24", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-01T00:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-01-rtx3080-mobile-tps/capture.json", "outputHash": "9e3ece081923a4d6d8bb0d2c467565d36f5e18d68ca5fd3ad2879e6662458f4a", "notes": "RTX 3080 Laptop (16GB), Windows, Ollama 0.24.0, sürücü 571.96, CUDA 12.8. Ollama etiketi gemma3:12b; capture dosyasındaki nicemleme alanı Q4_K_M. 5 koşu, ısınma koşusu atıldı, temperature 0, seed 42. Değer 5 koşunun medyanıdır (ortalama 43.32, p5 43.21, p95 43.42 tok/s); capture dosyası koşu başına değerleri değil bu özet istatistikleri yayımlar." }, { "id": "trendyol-llm-8b-v2-tps-m2max-q8-20260612", "modelSlug": "trendyol-llm-8b-v2", "dimension": "tps", "scores": [ 40.5, 40.64, 40.62, 40.6, 40.52 ], "sampleCount": 5, "hardware": "m2-max-32gb", "runtime": "ollama-0.30", "quantization": "q8_0", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-12T12:05:00Z", "submittedBy": "Eruo Fredoline", "status": "editorial-estimate", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://www.yerel.ai/acik-veri", "notes": "Mac Studio M2 Max (32GB birleşik bellek), macOS 26.5.1, Ollama 0.30.7 Metal. GGUF: mradermacher/Trendyol-LLM-8b-chat-v2.0-GGUF Q8_0. batch=1, 4096 ctx, temp 0, seed 42; 1 warmup atıldı, 5 ardışık run, varyans %0.15. Üretim 69 tokende EOS ile bitti (steady-state decode geçerli). Prefill_tps median 689 t/s (~1.2K token prompt, cache-bust), sıcak TTFT median 25 ms." }, { "id": "gemma-3-27b-tps-m3max-q4-20260612", "modelSlug": "gemma-3-27b-it", "dimension": "tps", "scores": [ 18.89, 18.8, 18.2, 17.49, 17.63 ], "sampleCount": 5, "hardware": "m3-max-128gb", "runtime": "ollama-0.30", "quantization": "q4_k_m", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-12T13:50:37Z", "submittedBy": "Eruo Fredoline", "status": "editorial-estimate", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://www.yerel.ai/acik-veri", "notes": "MacBook Pro M3 Max (128GB birleşik bellek, Mac15,9), macOS 26.5.1, Ollama 0.30.7 Metal. batch=1, 4096 ctx, 256 yeni token, temp 0, seed 42; 1 warmup atıldı, 5 ardışık run. Hafif ısıl düşüş gözlendi (18.89→17.63, laptop kasası). Prefill_tps ortalama 154 t/s (~1.2K token prompt, cache-bust), sıcak TTFT median 413 ms." }, { "id": "trendyol-llm-8b-v2-tps-m3max-q8-20260612", "modelSlug": "trendyol-llm-8b-v2", "dimension": "tps", "scores": [ 40.68, 40.6, 40.78, 40.57, 40.63 ], "sampleCount": 5, "hardware": "m3-max-128gb", "runtime": "ollama-0.30", "quantization": "q8_0", "contextLength": 4096, "seed": 42, "datasetVersion": "tps-2026.06", "submittedAt": "2026-06-12T14:41:35Z", "submittedBy": "Eruo Fredoline", "status": "editorial-estimate", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://www.yerel.ai/acik-veri", "notes": "MacBook Pro M3 Max (128GB birleşik bellek), macOS 26.5.1, Ollama 0.30.7 Metal. Soğuk kasa protokolü (10 dk soğutma, izole çalışma). batch=1, 4096 ctx, temp 0, seed 42; 1 warmup atıldı, 5 run, varyans %0.2. M2 Max ile istatistiksel olarak özdeş (40.58 vs 40.65) — iki çip de 400 GB/s, decode bant genişliği sınırlı. Prefill_tps ortalama 753 t/s, sıcak TTFT median 25 ms. Aynı gistte ısıl-kirlenmiş ilk deneme (20.6) şeffaflık için saklanıyor." }, { "id": "qwen-2-5-7b-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "qwen-2-5-7b", "dimension": "turkishmmlu-yuksel", "scores": [ 53.67 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T16:29:48Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-18-qwen-2-5-7b-turkishmmlu-yuksel/results.json", "outputHash": "a7a4104d03fdc488aa8754db2689dfea88b7277c798add9a6d92f5583530054b", "notes": "lm-evaluation-harness 0.4.12, task=turkishmmlu, --num_fewshot 5 --apply_chat_template (fewshot_as_multiturn), seed=42, dtype=bfloat16, batch_size auto. NVIDIA RTX 4090 24GB (driver 590.48.01), Vast.ai. 9 dersin makro ortalaması (100 soru/ders, 900 toplam). Tek tohum birinci-elden ölçüm; koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir. NOT alibayram protokolü; Cetvel'in raporladığı 47.6'dan chat-template multiturn + makro-ortalama farkıyla ayrılır." }, { "id": "turkish-llama-8b-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "turkish-llama-8b", "dimension": "turkishmmlu-yuksel", "scores": [ 48.56 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T16:37:23Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-18-turkish-llama-8b-turkishmmlu-yuksel/results.json", "outputHash": "1041115f1e3c831bc0589ba99ee1948f7fdb88200256a83e3f8a1ca2af9f6f9e", "notes": "ytu-ce-cosmos/Turkish-Llama-8b-Instruct-v0.1. lm-evaluation-harness 0.4.12, task=turkishmmlu, 5-shot, apply_chat_template (fewshot_as_multiturn), seed=42, dtype=bfloat16, batch_size auto. NVIDIA RTX 4090 24GB, Vast.ai. 9 dersin makro ortalaması. Birinci-elden; bu ölçümde Qwen2.5-7B (53.67) Türkçe-özel Cosmos-Llama-8b'yi geçiyor." }, { "id": "ytu-turkish-gemma-9b-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "ytu-turkish-gemma-9b", "dimension": "turkishmmlu-yuksel", "scores": [ 62.33 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:01:32Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-18-ytu-turkish-gemma-9b-turkishmmlu-yuksel/results.json", "outputHash": "7bc3bd416ee88edce28eb782c1f7d3bea51a9824cfa6fb26c4407efafb175e4b", "notes": "ytu-ce-cosmos/Turkish-Gemma-9b-v0.1. lm-evaluation-harness 0.4.12, task=turkishmmlu, 5-shot, apply_chat_template (fewshot_as_multiturn), seed=42, dtype=bfloat16, batch_size auto. NVIDIA RTX 4090 24GB, Vast.ai. 9 dersin makro ortalaması (900 soru). Birinci-elden — bu boyutta açık Türkçe modeller arasında lider (Qwen2.5-7B 53.67 ve Cosmos-Llama-8b 48.56 üstünde)." }, { "id": "turkcell-llm-7b-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "turkcell-llm-7b", "dimension": "turkishmmlu-yuksel", "scores": [ 32.78 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:24:14Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-06-18-turkcell-llm-7b-turkishmmlu-yuksel/results.json", "outputHash": "a91d21ad21efc25fbfd05863543aa6c556fa151d4b29f880f615a2c345406438", "notes": "TURKCELL/Turkcell-LLM-7b-v1. lm-evaluation-harness 0.4.12, task=turkishmmlu, 5-shot, apply_chat_template (fewshot_as_multiturn), seed=42, dtype=bfloat16, batch_size auto. NVIDIA RTX 4090 24GB, Vast.ai. 9 dersin makro ortalaması (900 soru). Birinci-elden ölçüm." }, { "id": "kumru-2b-turkishmmlu-yuksel-rtx3080laptop-20260708", "modelSlug": "kumru-2b", "dimension": "turkishmmlu-yuksel", "scores": [ 20.67 ], "sampleCount": 900, "hardware": "rtx-3080-laptop-16gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 8192, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-07-08T18:24:08Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/2026-07-08-kumru-2b-turkishmmlu-yuksel/results.json", "outputHash": "c483cc5d820f46ccd3ce7374bc06c038640e5eaca51e3dfc6893eae21ea98b79", "notes": "vngrs-ai/Kumru-2B (temel/base 2B, Mistral mimarisi). lm-evaluation-harness 0.4.12, task=turkishmmlu, --num_fewshot 5, seed=42, dtype=bfloat16, batch_size 8, sohbet şablonu YOK (temel model). NVIDIA RTX 3080 Laptop 16GB — yerel, birinci-elden ölçüm (bu PC'de koşuldu). 9 dersin makro ortalaması (100 soru/ders, 900 toplam) = %20.67. Küçük bir TEMEL model olarak 5-şıklı akademik MC'de rastgele (≈%20) seviyesinde; benzer 2B Kanarya-2B (%20.78) ile uyumlu. Loglikelihood MC ölçümü determinist (GPU-bağımsız); results.json + SHA-256 + pinli harness/komut ile yayımlanır." }, { "id": "turkish-gemma-9b-t1-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "turkish-gemma-9b-t1", "dimension": "turkishmmlu-yuksel", "scores": [ 55.67 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/ytu-ce-cosmos_Turkish-Gemma-9b-T1.json", "outputHash": "efde3fd406b174284ca75026eb7eb8ecebbf23ccaf44353629f1bf66895753f3", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "trendyol-llm-7b-chat-v4-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "trendyol-llm-7b-chat-v4", "dimension": "turkishmmlu-yuksel", "scores": [ 54.11 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/Trendyol_Trendyol-LLM-7B-chat-v4.1.0.json", "outputHash": "690c1c84e24121167d61361495ddf18373deff92dbee423c5df011dd913616ce", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "kizagan-e4b-turkish-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "kizagan-e4b-turkish", "dimension": "turkishmmlu-yuksel", "scores": [ 47.44 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/Tuguberk_Kizagan-E4B-Turkish-Agent-FunctionCalling-Hermes.json", "outputHash": "8dcabea3b0ccdc94dc19c8ce654c1d4796117fef44dd2e67c5452da8ab2bf865", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "qwen-2b-turkish-sft-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "qwen-2b-turkish-sft", "dimension": "turkishmmlu-yuksel", "scores": [ 39.78 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/Tuguberk_Qwen3.5-2B-Turkish-SFT.json", "outputHash": "7fa19cc7b44b1a26b730d52dfb235bf96125ab1ea3cae92ef26dbf2401a3791c", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "openhermes-strix-mistral-7b-tr-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "openhermes-strix-mistral-7b-tr", "dimension": "turkishmmlu-yuksel", "scores": [ 27.67 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/sayhan_OpenHermes-2.5-Strix-Philosophy-Mistral-7B-LoRA.json", "outputHash": "dea003a28de2f1aba20b1c3e6563b00652ea302ba1fe6190c6a08e420e6fb0e2", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "mam-turkish-790m-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "mam-turkish-790m", "dimension": "turkishmmlu-yuksel", "scores": [ 23.89 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/yusuf229_mam-turkish-790m.json", "outputHash": "aac51a9682f158deddb760ed61d8f024234ac406e02448a13755fe34c6848620", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "kanarya-750m-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "kanarya-750m", "dimension": "turkishmmlu-yuksel", "scores": [ 21 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/asafaya_kanarya-750m.json", "outputHash": "7dda249fbf4b0fa962a8cc0ccb8fd471fdb5762456242f19fe858d94d031c281", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "kanarya-2b-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "kanarya-2b", "dimension": "turkishmmlu-yuksel", "scores": [ 20.78 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/asafaya_kanarya-2b.json", "outputHash": "61d8af6502c058cca66229ae4ec8d0a6b906f17337c8d32f433136b1e441c8b8", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "trendyol-llm-7b-chat-v01-turkishmmlu-yuksel-rtx4090-20260618", "modelSlug": "trendyol-llm-7b-chat-v01", "dimension": "turkishmmlu-yuksel", "scores": [ 18.56 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-18T18:30:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch-2026-06-18/Trendyol_Trendyol-LLM-7b-chat-v0.1.json", "outputHash": "edb9b6ab293d0c0ca6bc23a6f2a3e151bf1e7ba2560493d5cc2c245110fb792f", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "trendyol-llm-asure-12b-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "trendyol-llm-asure-12b", "dimension": "turkishmmlu-yuksel", "scores": [ 62.22 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/Trendyol_Trendyol-LLM-Asure-12B.json", "outputHash": "e7bcbd77c279b38b3e3ee4a473f9851cc432716b2674df6661009fe6d23a01c4", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "cosmos-tr-culture-9b-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "cosmos-tr-culture-9b", "dimension": "turkishmmlu-yuksel", "scores": [ 55.44 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/ahmet-erman_cosmos-turkish-culture-veri_1-full_epoch.json", "outputHash": "d59f912320a773b1816f9f3b25bba6a5b56a42e3756200d7b56cadf0d59784e6", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "qwen25-7b-tr-culture-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "qwen25-7b-tr-culture", "dimension": "turkishmmlu-yuksel", "scores": [ 54 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/ahmet-erman_Qwen2.5-7B-turkish-culture-veri_1-full_epoch.json", "outputHash": "8c8e815383b2425fca9bb9a3d651356a0bcb5477bf41eebbb26b920cdce6a554", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "kizagan-e4b-reasoning-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "kizagan-e4b-reasoning", "dimension": "turkishmmlu-yuksel", "scores": [ 51.22 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/AlicanKiraz0_Kizagan-E4B-Turkish-Reasoning-Model.json", "outputHash": "8491356ea0c486fa613650d3c00af105305d98c0bd7399bc573e9ec84ec26c78", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "wiroai-turkish-9b-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "wiroai-turkish-9b", "dimension": "turkishmmlu-yuksel", "scores": [ 47.89 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/WiroAI_wiroai-turkish-llm-9b.json", "outputHash": "f2aaa9a5c123bea5ed9b3fcabc3f25b9385305b69376aa5021e4125aa2b756e2", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "trendyol-llm-8b-v2-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "trendyol-llm-8b-v2", "dimension": "turkishmmlu-yuksel", "scores": [ 46.11 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/Trendyol_Llama-3-Trendyol-LLM-8b-chat-v2.0.json", "outputHash": "1aa4cfd37c108dbd8e5c9069778b92210eeb712edb1e974c7329be0bf24784cd", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "turkish-gemma-4b-scout-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "turkish-gemma-4b-scout", "dimension": "turkishmmlu-yuksel", "scores": [ 44.78 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/ytu-ce-cosmos_Turkish-Gemma-4b-T1-Scout.json", "outputHash": "e497bc00384d4f5cba9598faaf944589476645d4b399dd01e9485ca7cd946fbf", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "llama3-8b-tr-culture-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "llama3-8b-tr-culture", "dimension": "turkishmmlu-yuksel", "scores": [ 44.67 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/ahmet-erman_LLama-3-8B-turkish-culture-veri_1-full_epoch.json", "outputHash": "dcc68c653df1d238f7c30e4b00d5003d2dae3c4a25b48d6826776fb97195009b", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "turkish-llama-8b-base-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "turkish-llama-8b-base", "dimension": "turkishmmlu-yuksel", "scores": [ 44.67 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/ytu-ce-cosmos_Turkish-Llama-8b-v0.1.json", "outputHash": "1dc88def0f3cedee423082b5a40edf4e437312862157e76c37f42903edb8629d", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "trendyol-llm-7b-base-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "trendyol-llm-7b-base", "dimension": "turkishmmlu-yuksel", "scores": [ 22.67 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/Trendyol_Trendyol-LLM-7b-base-v1.0.json", "outputHash": "e3cabb1eb360bfd24a37072b5b7cd1e4f1da513ffee653b3fbcd6391bd5b8426", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "trendyol-llm-8b-t1-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "trendyol-llm-8b-t1", "dimension": "turkishmmlu-yuksel", "scores": [ 21.33 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/Trendyol_Trendyol-LLM-8B-T1.json", "outputHash": "fce5fd80892b34e66236e290e1f638ff9991a36fcda0e3035f428e84cb9d9a62", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "kumru-2b-reasoning-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "kumru-2b-reasoning", "dimension": "turkishmmlu-yuksel", "scores": [ 21.22 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/DogaOytc_kumru-2b-turkish-reasoning-grpo.json", "outputHash": "b46e3259120fee7c138b9abda7db59abf2afb480d705f5cc4cbbe9fff7916642", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "kara-kumru-2b-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "kara-kumru-2b", "dimension": "turkishmmlu-yuksel", "scores": [ 19.89 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/AlicanKiraz0_Kara-Kumru-v1.0-2B.json", "outputHash": "8a3764c77a887e6b90671be0818a5593589b91326a1afd925508fda20098bb4a", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "tr-qwen25-05b-sft-turkishmmlu-yuksel-rtx4090-20260619", "modelSlug": "tr-qwen25-05b-sft", "dimension": "turkishmmlu-yuksel", "scores": [ 18.22 ], "sampleCount": 900, "hardware": "rtx-4090-24gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "turkishmmlu-yuksel-2026.06", "submittedAt": "2026-06-19T07:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/turkishmmlu-yuksel-batch2-2026-06-19/ytu-ce-cosmos_tr-Qwen2.5-0.5B-SFT-v1.json", "outputHash": "2034426d62edd45975750b87ab4448933d171bd2694dfac107aafb77f37c0361", "notes": "lm-evaluation-harness 0.4.12 turkishmmlu, 5-shot, seed 42, bf16, RTX 4090 (Vast.ai); sohbet şablonu modelde mevcutsa uygulandı. 9 dersin makro ortalaması, 900 soru. Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "trendyol-llm-asure-12b-belebele-tr-rtx5090-20260619", "modelSlug": "trendyol-llm-asure-12b", "dimension": "belebele-tr", "scores": [ 86.56 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/Trendyol__Trendyol-LLM-Asure-12B.json", "outputHash": "f19b9d98a7880922876baa931fcdd8845c547b40324608ccd517ead2886c5283", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "ytu-turkish-gemma-9b-belebele-tr-rtx5090-20260619", "modelSlug": "ytu-turkish-gemma-9b", "dimension": "belebele-tr", "scores": [ 85.89 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/ytu-ce-cosmos__Turkish-Gemma-9b-v0.1.json", "outputHash": "0d30f3dd05d1ff2130f409ca19ea0c7db8fd8e4655b2070fc8227670131dbb34", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "turkish-gemma-9b-t1-belebele-tr-rtx5090-20260619", "modelSlug": "turkish-gemma-9b-t1", "dimension": "belebele-tr", "scores": [ 83.89 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/ytu-ce-cosmos__Turkish-Gemma-9b-T1.json", "outputHash": "9cc10b20c926ad930959bd160c955e3c06da47dcb3323a34f0704834afc651e5", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "cosmos-tr-culture-9b-belebele-tr-rtx5090-20260619", "modelSlug": "cosmos-tr-culture-9b", "dimension": "belebele-tr", "scores": [ 82.11 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/ahmet-erman__cosmos-turkish-culture-veri_1-full_epoch.json", "outputHash": "558069a425fbea926e197e64a047132c5bfe6bbafcf857d42267bb862f7c2b99", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "trendyol-llm-7b-chat-v4-belebele-tr-rtx5090-20260619", "modelSlug": "trendyol-llm-7b-chat-v4", "dimension": "belebele-tr", "scores": [ 77.89 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/Trendyol__Trendyol-LLM-7B-chat-v4.1.0.json", "outputHash": "42024a897fb3d6d452166817312b074787e06c2be468467021edb751e5164e87", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "wiroai-turkish-9b-belebele-tr-rtx5090-20260619", "modelSlug": "wiroai-turkish-9b", "dimension": "belebele-tr", "scores": [ 82.56 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/WiroAI__wiroai-turkish-llm-9b.json", "outputHash": "83265f25c5235f0e7257fba28bed509802d0a877c1dd516b584e2ccaace71818", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "qwen-2-5-7b-belebele-tr-rtx5090-20260619", "modelSlug": "qwen-2-5-7b", "dimension": "belebele-tr", "scores": [ 78.78 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/Qwen__Qwen2.5-7B-Instruct.json", "outputHash": "c7a15ee762cc3a37d01cfc174878090ced1319b22b3d0e90cf5d22a542b9c4d9", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "kizagan-e4b-reasoning-belebele-tr-rtx5090-20260619", "modelSlug": "kizagan-e4b-reasoning", "dimension": "belebele-tr", "scores": [ 76 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/AlicanKiraz0__Kizagan-E4B-Turkish-Reasoning-Model.json", "outputHash": "957b5dcd9632398322b8c88b8d08ed30d766169c04f86869a5597b7e9c2bb964", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "qwen25-7b-tr-culture-belebele-tr-rtx5090-20260619", "modelSlug": "qwen25-7b-tr-culture", "dimension": "belebele-tr", "scores": [ 73.56 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/ahmet-erman__Qwen2.5-7B-turkish-culture-veri_1-full_epoch.json", "outputHash": "732a988ea298ebba31422217af545e6d12b0a97c9e7a55d058bb9026aa696ca3", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "kizagan-e4b-turkish-belebele-tr-rtx5090-20260619", "modelSlug": "kizagan-e4b-turkish", "dimension": "belebele-tr", "scores": [ 71.44 ], "sampleCount": 900, "hardware": "rtx-5090-32gb", "runtime": "lm-eval-hf", "quantization": "bf16", "contextLength": 4096, "seed": 42, "datasetVersion": "belebele-tur-Latn-2026.06", "submittedAt": "2026-06-19T11:00:00Z", "submittedBy": "Eruo Fredoline", "status": "editor-pick", "verifications": 1, "measurement": "direct", "sourceUrl": "https://www.yerel.ai/api/runs/belebele-tr-2026-06-19/Tuguberk__Kizagan-E4B-Turkish-Agent-FunctionCalling-Hermes.json", "outputHash": "5ae5f6b66ebf94597de360fdfd426540f849bf1b7cab0defc27079aa33be793b", "notes": "lm-evaluation-harness belebele_tur_Latn, 5-shot, apply_chat_template, seed 42, bf16, RTX 5090 (Vast.ai). Meta Belebele Türkçe okuduğunu-anlama (900 soru). Her koşunun results.json'u (per-subject aggregate sonuçlar) + SHA-256 çıktı karması + pinli komut & harness sürümü ile yayımlanır; bağımsız biri aynı kurulumda bayt-birebir tekrar üretebilir." }, { "id": "ytu-turkish-gemma-9b-tr-mmlu-ref-alibayram", "modelSlug": "ytu-turkish-gemma-9b", "dimension": "tr-mmlu", "scores": [ 70.34 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"alibayram/turkish-gemma-9b-v0.1\". GGUF, Ollama tablo değeri. YTÜ Turkish-Gemma 9B v0.1 (alibayram tarafından Ollama’a yayınlandı). En iyi Türkçe ince-ayar modeli." }, { "id": "turkish-llama-8b-tr-mmlu-ref-alibayram", "modelSlug": "turkish-llama-8b", "dimension": "tr-mmlu", "scores": [ 54.94 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"ytu-ce-cosmos-Turkish-Llama-8b-DPO-v0-1\". GGUF, Ollama tablo değeri. Liderlik tablosunda DPO-v0.1 varyantının alıntılanan değeri; katalogdaki Instruct-v0.1 ile aynı model ailesi." }, { "id": "trendyol-llm-8b-v2-tr-mmlu-ref-alibayram", "modelSlug": "trendyol-llm-8b-v2", "dimension": "tr-mmlu", "scores": [ 27.58 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"Trendyol-LLM-8b-chat-v2.0\". GGUF, Ollama tablo değeri. Türkçe ince-ayarlı olmasına rağmen genel modellerin altında — TR-MMLU bilgi tablo değerinde." }, { "id": "kocdigital-llm-8b-tr-mmlu-ref-alibayram", "modelSlug": "kocdigital-llm-8b", "dimension": "tr-mmlu", "scores": [ 27.47 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"Kocdigital-LLM-8B\". GGUF, Ollama tablo değeri." }, { "id": "turkcell-llm-7b-tr-mmlu-ref-alibayram", "modelSlug": "turkcell-llm-7b", "dimension": "tr-mmlu", "scores": [ 19.5 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"Turkcell-LLM-7b-v1\". GGUF, Ollama tablo değeri." }, { "id": "gpt-4o-tr-mmlu-ref-alibayram", "modelSlug": "gpt-4o", "dimension": "tr-mmlu", "scores": [ 84.84 ], "sampleCount": 6200, "hardware": "api", "runtime": "api-direct", "quantization": "api", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"gpt-4o\". API (tam hassasiyet) tablo değeri." }, { "id": "claude-3-5-sonnet-tr-mmlu-ref-alibayram", "modelSlug": "claude-3-5-sonnet", "dimension": "tr-mmlu", "scores": [ 84.4 ], "sampleCount": 6200, "hardware": "api", "runtime": "api-direct", "quantization": "api", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"claude-3-5-sonnet-20240620\". API (tam hassasiyet) tablo değeri." }, { "id": "llama-3-3-70b-tr-mmlu-ref-alibayram", "modelSlug": "llama-3-3-70b", "dimension": "tr-mmlu", "scores": [ 79.42 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"llama3.3:latest (70.6B)\". GGUF, Ollama tablo değeri." }, { "id": "gemini-1-5-pro-tr-mmlu-ref-alibayram", "modelSlug": "gemini-1-5-pro", "dimension": "tr-mmlu", "scores": [ 76.74 ], "sampleCount": 6200, "hardware": "api", "runtime": "api-direct", "quantization": "api", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"gemini-1.5-pro\". API (tam hassasiyet) tablo değeri." }, { "id": "gemma-3-27b-it-tr-mmlu-ref-alibayram", "modelSlug": "gemma-3-27b-it", "dimension": "tr-mmlu", "scores": [ 75.06 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"gemma3:27b\". GGUF, Ollama tablo değeri." }, { "id": "gemma-2-27b-tr-mmlu-ref-alibayram", "modelSlug": "gemma-2-27b", "dimension": "tr-mmlu", "scores": [ 72.1 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_0", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"gemma2:27b\". GGUF, Ollama tablo değeri." }, { "id": "gemma-3-12b-it-tr-mmlu-ref-alibayram", "modelSlug": "gemma-3-12b-it", "dimension": "tr-mmlu", "scores": [ 70.74 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"gemma3:12b\". GGUF, Ollama tablo değeri." }, { "id": "aya-expanse-32b-tr-mmlu-ref-alibayram", "modelSlug": "aya-expanse-32b", "dimension": "tr-mmlu", "scores": [ 70.66 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"aya-expanse:32b\". GGUF, Ollama tablo değeri." }, { "id": "phi-4-tr-mmlu-ref-alibayram", "modelSlug": "phi-4", "dimension": "tr-mmlu", "scores": [ 65.52 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"phi4:latest (14.7B)\". GGUF, Ollama tablo değeri." }, { "id": "qwen-2-5-7b-tr-mmlu-ref-alibayram", "modelSlug": "qwen-2-5-7b", "dimension": "tr-mmlu", "scores": [ 61.68 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"qwen2.5:latest (7.6B)\". GGUF, Ollama tablo değeri." }, { "id": "gemma-3-4b-it-tr-mmlu-ref-alibayram", "modelSlug": "gemma-3-4b-it", "dimension": "tr-mmlu", "scores": [ 42.74 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"gemma3:4b\". GGUF, Ollama tablo değeri." }, { "id": "gemma-3-1b-it-tr-mmlu-ref-alibayram", "modelSlug": "gemma-3-1b-it", "dimension": "tr-mmlu", "scores": [ 42.74 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_k_m", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"gemma3:1b\". GGUF, Ollama tablo değeri." }, { "id": "llama-3-1-8b-tr-mmlu-ref-alibayram", "modelSlug": "llama-3-1-8b", "dimension": "tr-mmlu", "scores": [ 35.11 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_0", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"llama3.1:latest (8B)\". GGUF, Ollama tablo değeri. Liderlik tablosundaki bu run düşük görünüyor (olası kurulum/parse sorunu); kaynaktaki değer aynen aktarıldı." }, { "id": "mistral-7b-v0-3-tr-mmlu-ref-alibayram", "modelSlug": "mistral-7b-v0-3", "dimension": "tr-mmlu", "scores": [ 32.73 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_0", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"mistral:7b\". GGUF, Ollama tablo değeri." }, { "id": "phi-3-5-mini-tr-mmlu-ref-alibayram", "modelSlug": "phi-3-5-mini", "dimension": "tr-mmlu", "scores": [ 29.37 ], "sampleCount": 6200, "hardware": "reference-leaderboard", "runtime": "ollama-0.6", "quantization": "q4_0", "contextLength": 4096, "seed": 0, "datasetVersion": "turkish-mmlu-alibayram-2026.06", "submittedAt": "2026-06-14T00:00:00Z", "submittedBy": "alibayram TurkishMMLU Leaderboard", "status": "community", "verifications": 1, "measurement": "aggregated", "sourceUrl": "https://huggingface.co/spaces/alibayram/turkish_mmlu_leaderboard", "notes": "Alıntı (alibayram liderlik tablosu): alibayram TurkishMMLU liderlik tablosu — kayıt: \"phi3.5:latest\". GGUF, Ollama tablo değeri." } ] }