NoisyBench - Hard Negative Distractors: leaderboard
Metric: Harmonic Mean Score (%; 11 datasets' pass@k, pass^k for TauBench; a task-specific hard-negative distractor added). Source: arxiv.org. Saturation forecast: Around December 2026. 7 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-OSS-120B (High) | 50.2 |
| 2 | Gemini 2.5 Pro | 48 |
| 3 | DeepSeek R1 0528 | 47.6 |
| 4 | Gemini 2.5 Flash | 45.6 |
| 5 | Qwen 3 30B A3B 2507 (Thinking) | 35 |
| 6 | Qwen 3 4B 2507 (Thinking) | 32.7 |
| 7 | DeepSeek R1 Distill Llama 8B | 6.3 |
Interactive version: theaggregate.ai/benchmark?slug=noisybench-hard-negative-distractors · How It Works · Data refreshed daily, snapshot 2026-09-25.