IFEval-TR - Translated (strict): leaderboard
Measures verifiable instruction following in Turkish on IFEval-TR, a Turkish adaptation of Google's IFEval covering 910 examples and 25 constraint types across length, keyword and letter frequency, formatting, casing, punctuation and start/end markers. Every constraint is checked by a deterministic verifier; no LLM judge is involved. This series is the Translated set: 498 examples rendered from the English IFEval, with language-dependent parameters also translated. Strict scoring requires exact string matching with no morphological relaxation, which penalises morphologically equivalent Turkish responses. Scores are prompt-level accuracy. This series reports IFEval-TR - Translated (strict) using Accuracy (%).
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Around February 2027. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.5 27B FP8 | 69.1 | #171 |
| 2 | Qwen 3.8 27B | 67.47 | #123 |
| 3 | Qwen 3.5 9B | 64.79 | #391 |
| 4 | GPT-OSS-120B | 62.5 | #346 |
| 5 | GPT-OSS-20B | 58.37 | #523 |
| 6 | Gemma 3 12B (IT) | 56.43 | #672 |
| 7 | Gemma 3 27B (IT) | 56.22 | #528 |
| 8 | GPT-4o Mini | 53.8 | #598 |
| 9 | Gemma 3 4B (IT) | 47.99 | #997 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ifeval-tr-translated-strict · How It Works · Data refreshed daily, snapshot 2026-11-05.