BigCodeBench-Hard: leaderboard
BigCodeBench-Hard evaluates code generation on the harder BigCodeBench subset, reporting pass@1 in complete and instruct settings.
Metric: Instruct pass@1 (self-reported). Source: benchmarklist.com. Status: saturated. 20 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 Mini (2025-01-31) | 33.1 |
| 2 | Claude 3.7 Sonnet (20250219) | 32.4 |
| 3 | O1 (2024-12-17) | 32.4 |
| 4 | GPT-4.1 Mini | 31.8 |
| 5 | GPT-4.1 | 31.8 |
| 6 | Grok 3 Mini Beta | 31.1 |
| 7 | Gemini 2.5 Pro (Preview 03-25) | 29.7 |
| 8 | DeepSeek R1 | 29.7 |
| 9 | GPT-4 Turbo | 29.1 |
| 10 | Qwen 2.5 Max | 29.1 |
| 11 | Llama 3.3 70B Instruct | 28.4 |
| 12 | GPT-4.1 Nano | 28.4 |
| 13 | DeepSeek V3 | 28.4 |
| 14 | GPT-4o (2024-11-20) | 27.7 |
| 15 | Qwen 2.5 Coder 32B Instruct | 27.7 |
Interactive version: theaggregate.ai/benchmark?slug=bigcodebench-hard · How It Works · Data refreshed daily, snapshot 2026-09-05.