InfiBench: leaderboard
InfiBench — freeform code question answering on real developer questions (Stack Overflow style) with rubric-based automatic grading.
Metric: Score (%). Source: infi-coder.github.io. Saturation forecast: Estimated already saturated. 106 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4 (0613) | 70.64 | #491 |
| 2 | GPT-4 Turbo (Preview) | 68.42 | #397 |
| 3 | GPT-4o (2024-05-13) | 66.19 | #325 |
| 4 | Phind-CodeLlama-34B-v2 | 59 | #1589 |
| 5 | Mistral Large | 58.22 | #710 |
| 6 | GPT-3.5 Turbo (0613) | 56.47 | #1118 |
| 7 | Mistral Small | 55.62 | #740 |
| 8 | Qwen-72B | 55.34 | #779 |
| 9 | Qwen-72B-Chat | 52.97 | #843 |
| 10 | Magicoder-S-CL-7B | 52.71 | #1489 |
| 11 | Magicoder-S-DS-6.7B | 51.46 | #1256 |
| 12 | Yi Chat 34B | 49.58 | #976 |
| 13 | zephyr-7B-beta | 46.31 | #1330 |
| 14 | Qwen-14B | 43.69 | #1171 |
| 15 | Qwen-14B-Chat | 43.49 | #1169 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=infibench · How It Works · Data refreshed daily, snapshot 2026-10-20.