MEGA-Bench Task - Code Programming Extremely Hard — leaderboard
Metric: Task Score (%). Source: huggingface.co. 44 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude 3.5 Sonnet (20241022) | 9.4 |
| 2 | Ovis2-8B | 6.2 |
| 3 | Gemini 1.5 Pro (002) | 6.2 |
| 4 | Claude 3.5 Sonnet (20240620) | 6.2 |
| 5 | Gemini 2.0 Flash (Preview) | 6.2 |
| 6 | Gemma 3 27B (IT) | 3.1 |
| 7 | Llama 4 Scout Base | 3.1 |
| 8 | Gemini 2.5 Pro | 0 |
| 9 | GPT-4o | 0 |
| 10 | Gemma 3 12B (IT) | 0 |
| 11 | Gemma 3 4B (IT) | 0 |
| 12 | InternVL3-78B | 0 |
| 13 | InternVL2-8B | 0 |
| 14 | InternVL3-8B | 0 |
| 15 | Qwen 2 VL 72B | 0 |
Interactive version: theaggregate.ai/benchmark?slug=mega-bench-task-code-programming-extremely-hard · How the rankings work · Data refreshed daily, snapshot 2026-07-22.