StressEval - Text: leaderboard

StressEval (2026) builds Dynamic-OneEval by synthesizing new knowledge-intensive questions from GPT-5.2 failure cases on HotpotQA, ComplexWebQuestions and Wikipedia table question sets, keeping the difficulty factor that caused each failure. This board is exact-match accuracy in percent on the text-reasoning splits (knowledge stress and reasoning stress combined); higher is better.

Metric: Exact-match accuracy (%) on the text-reasoning splits (knowledge stress and reasoning stress, 856 instances seeded from HotpotQA, instance-weighted mean) of Dynamic-OneEval (StressEval instances synthesized by GPT-5.2 from its own failure cases on HotpotQA, ComplexWebQuestions and WikiTableQuestions, gated for answerability and consistency), one unified prompt template for every model; higher is better. Source: arxiv.org. Saturation forecast: Around July 2028. 9 models tracked. Updated 7 Oct 2026.

Top models

#ModelScoreOverall rank
1Gemini 3 Pro42.2#145
2Claude Sonnet 4.527.8#261
3Seed-1.625.1#469
4Llama 3.1 70B21.3#1499
5QwQ-32B20.5#907
6Qwen 3 235B A22B18.9#628
7Qwen 2.5 72B17.8#1188
8Llama 3.1 8B17.6#5196
9DeepSeek V3.214.8#419

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=stresseval-text · How It Works · Data refreshed daily, snapshot 2026-11-21.