Vals AI Vibe Code Bench 1-100: leaderboard
Vibe Code Bench 1-100 (Vals AI): starting from a fully passing Vibe Code Bench v1.1 web app, models extend it through up to ten dependent plain-English product requests, keeping code and conversation across turns. Each iteration runs the new request's tests plus regression tests for earlier behaviour, and the first incomplete iteration ends the attempt.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around 2030. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Muse Wonder Full (Max) | 33.98 | #12 (Muse Wonder Full) |
| 2 | Claude Opus 5.5 (xHigh) | 30.36 | #3 (Claude Opus 5.5) |
| 3 | Claude Opus 5 (Max) | 28.53 | #16 (Claude Opus 5) |
| 4 | Claude Fable 5.1 (Max) | 28 | #10 (Claude Fable 5.1) |
| 5 | GPT-6 Astra (Max) | 27.64 | #6 (GPT-6 Astra) |
| 6 | Muse Wonder (Max) | 26.55 | #18 (Muse Wonder) |
| 7 | GPT-5.6 Luna (Max) | 22.59 | #79 (GPT-5.6 Luna) |
| 8 | Gemini 4 Argon (High) | 21.2 | #9 (Gemini 4 Argon) |
| 9 | Muse Spark 1.3 (Max) | 20.46 | #31 (Muse Spark 1.3) |
| 10 | Claude Haiku 5.5 (Max) | 20.41 | #43 (Claude Haiku 5.5) |
| 11 | GPT-5.6 Sol (Max) | 20.01 | #22 (GPT-5.6 Sol) |
| 12 | GLM-5.3 (Max) | 19.99 | #44 (GLM-5.3) |
| 13 | Gemini 3.8 Flash (High) | 18.77 | #30 (Gemini 3.8 Flash) |
| 14 | Kimi K3 (Max) | 18.24 | #35 (Kimi K3) |
| 15 | DeepSeek V4 Pro (0813) (Max) | 17.53 | #58 (DeepSeek V4 Pro (0813)) |
No result here: #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-vibe-code-bench-1-100 · How It Works · Data refreshed daily, snapshot 2026-11-05.