LiveCodeBench
Code · 2024-03-12
LiveCodeBench evaluates code generation on newly released competitive-programming problems to limit contamination. Scores report pass@1 accuracy in percentage points.
Top models (higher is better)
| Model | Score |
|---|---|
| Gemini 2.5 Deep Think | 87.6 |
| GPT-5 Mini | 80.4 |
| Grok 4 | 79.0 |
| Grok 4.1 Fast | 76.5 |
| O4 Mini | 75.8 |
| Gemini 2.5 Pro Preview 05-06 | 75.6 |
| o3 | 72.0 |
| Gemini 3.1 Flash-Lite Preview | 72.0 |
| Gemini 2.5 Flash Preview (09-2025) | 71.7 |
| Grok 3 | 70.6 |
| DeepSeek-R1-0528 | 70.5 |
| Gemini 2.5 Pro Experimental 03-25 | 70.4 |
| DeepSeek-R1 | 64.3 |
| Gemini 2.5 Flash | 63.9 |
| Gemini 2.5 Flash Preview 04-17 | 63.5 |
| Gemini 2.5 Flash-Lite Preview (09-2025) | 58.4 |
| Haiku 4.5 | 53.2 |
| Opus 4 | 51.1 |
| Sonnet 4 | 48.9 |
| Gemini 2.5 Flash-Lite | 34.3 |
| Gemini 1.5 Pro | 34.2 |
| Gemini 1.5 Flash | 30.7 |
| Gemini 2.0 Flash-Lite | 28.9 |