FireBench - Positive Content Requirements: leaderboard
Metric: Share of the 200 positive content requirements samples (%) in which the model performs the expected action: including every mandatory element added to Arena-Hard 2.0 programming and writing prompts, judged against rubrics by GPT-4.1; FireBench enterprise and API instruction following; higher is better. Source: arxiv.org. Saturation forecast: Not forecast. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4.1 | 94.5 | #240 |
| 2 | GPT-OSS-120B | 91.5 | #330 |
| 3 | Kimi K2 Instruct (0905) | 90 | #247 |
| 4 | Qwen 3 235B A22B 2507 Instruct | 87 | #291 |
| 5 | GPT-5.1 (Medium) | 86.5 | #131 (GPT-5.1) |
| 6 | Qwen 3 235B A22B 2507 (Thinking) | 85.5 | #253 (Qwen 3 235B A22B 2507) |
| 7 | DeepSeek V3.1 Terminus | 84 | #212 |
| 8 | Kimi K2 (Thinking) | 81 | #236 (Kimi K2) |
| 9 | Llama 4 Maverick Instruct | 80.5 | #439 |
| 10 | Claude Sonnet 4.5 | 79 | #138 |
| 11 | GPT-5.1 Instant | 70.5 | #277 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=firebench-positive-content-requirements · How It Works · Data refreshed daily, snapshot 2026-10-11.