Vals AI Vibe Code Bench 1-100 v1.1: leaderboard

Vibe Code Bench1-100v1.1 — OpenHands agents extend50working web apps through dependent product requests. Mean completed fraction of each request arc(%),not full-arc pass rate. Opus5.5headline30.95includes12task fallback work by Opus4.8/Opus5 and is retained as a composite;pre-fallback23.55is a separate diagnostic.

Metric: Mean completed fraction of dependent request arcs (%). Source: www.vals.ai. Saturation forecast: Around August 2028. 8 models tracked. Updated 15 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Gemini 4 CarbonHigh49.73#2
2Claude Opus 5.5VCB run: harness=OpenHands, fallback=Claude Opus 4.8 or Claude Opus 5, edition=1.130.95#5
3GLM-5.3Max29.75#68
4MiMo-V2.6-Pro26.41#76
5Grok 4.7xHigh26.35#47
6DeepSeek V4.1 Flash (1017)Max26.15#41
7MiMo-V2.6-Flash24.5#155
8MiniMax-M3.1-FlashMax19.83#109

No result here: #1 Claude Fable 5.2, #6 GPT-6 Astra, #10 GPT-6.1 Sol.

Interactive version: theaggregate.ai/benchmark?slug=vals-ai-vibe-code-bench-1-100-v1-1 · How It Works · Data refreshed daily, snapshot 2026-11-16.