Vals AI LiveCodeBench: leaderboard
Contest programming problems from LeetCode, AtCoder and Codeforces: the model reads a statement with sample tests and writes a Python solution. The newest LiveCodeBench release, v6, holds over a thousand problems collected between May 2023 and 2025, and Vals AI scores accuracy by running hidden test cases. Vals has archived this board and no longer runs new model releases on it; higher is better.
Metric: Accuracy (%). Source: www.vals.ai. Saturation forecast: Around January 2027. 143 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5.1 (Max) | 90.52 | #9 (Claude Fable 5.1) |
| 2 | Claude Fable 5 (Max) | 89.78 | #16 (Claude Fable 5) |
| 3 | Gemini 3.8 Flash (High) | 89.48 | #21 (Gemini 3.8 Flash) |
| 4 | Claude Opus 5 (Max) | 89.03 | #12 (Claude Opus 5) |
| 5 | Gemini 3.7 Flash (High) | 88.65 | #23 (Gemini 3.7 Flash) |
| 6 | Gemini 3.1 Pro (Preview) (High) | 88.48 | #58 (Gemini 3.1 Pro (Preview)) |
| 7 | Grok 4.6 (High) | 88.22 | #26 (Grok 4.6) |
| 8 | Gemini 3.6 Flash (High) | 88.08 | #51 (Gemini 3.6 Flash) |
| 9 | GPT-5.2 Codex | 87.99 | #95 |
| 10 | Qwen 3.8 Max | 87.85 | #35 |
| 11 | Claude Opus 4.8 (Max) | 87.82 | #37 (Claude Opus 4.8) |
| 12 | Gemini 3.5 Flash (High) | 87.6 | #60 (Gemini 3.5 Flash) |
| 13 | DeepSeek V4 Pro (0813) (Max) | 87.53 | #53 (DeepSeek V4 Pro (0813)) |
| 14 | DeepSeek V4 Pro (Max) | 87.48 | #102 (DeepSeek V4 Pro) |
| 15 | Grok 4.5 (High) | 87.35 | #48 (Grok 4.5) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=vals-ai-livecodebench · How It Works · Data refreshed daily, snapshot 2026-10-20.