Evolving Requirements - Full StrictPass: leaderboard

61 frozen repository tasks, 207 requirement states. ClaudeCode2.1.74, 200ktoken context, high reasoning, provider-default sampling, network-disabled. Full one complete session; Evo one response/state; Evo+Comp initial plus<=2completion revisions/state; Hy4-Preview high feedbackgenerator20ktokens and<=2publicobservations/message. StrictPass allchecks; NodeCredit partialstate credit; FDAC feedback-discounted adaptivecorrectness. Pooled row withheld. Metric: Native score (%). Source: https://arxiv.org/html/2610.35488v1

Metric: Native score (%). Source: arxiv.org. Saturation forecast: Around January 2028. 12 models tracked. Updated 17 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1GPT-6 AstraHigh59#8
2GPT-5.6 SolHigh52.5#33
3Hy4 previewHigh47.5#75
4GLM-5.3High44.3#73
5DeepSeek V4.1 FlashHigh42.6#69
6Kimi K3High39.3#59
7Qwen 3.8 FlashHigh39.3#128
8DeepSeek V4 FlashHigh37.7#265
9DeepSeek V4 ProHigh36.1#198
10GLM-5.3 FlashHigh36.1#136
11Grok 4.6High27.9#47
12Hy3High21.3#211

No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.

Interactive version: theaggregate.ai/benchmark?slug=evolving-requirements-full-strictpass · How It Works · Data refreshed daily, snapshot 2026-11-17.