StressEval: leaderboard
Static benchmarks for LLMs are increasingly compromised by contamination and overfitting especially on knowledge intensive reasoning tasks While recent dynamic benchmarks can alleviate staleness they often increase difficulty at the expense of answerability and controllability In this paper we propose StressEval a failure driven data synthesis framework that turns observed model failures into dyna
Metric: Overall (self-reported). Source: benchmarklist.com. Saturation forecast: Around February 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 46.4 | #80 |
| 2 | Claude Sonnet 4.5 | 37.4 | #142 |
| 3 | QwQ-32B | 34.6 | #421 |
| 4 | Qwen 2.5 72B | 33.9 | #499 |
| 5 | Llama 3.1 70B | 31.7 | #592 |
| 6 | Qwen 3 235B A22B | 28.7 | #313 |
| 7 | Seed-1.6 | 28.1 | #265 |
| 8 | DeepSeek V3.2 | 27.7 | #205 |
| 9 | Llama 3.1 8B | 24.7 | #1151 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=stresseval · How It Works · Data refreshed daily, snapshot 2026-10-20.