CodeARC - Annotated: leaderboard
The CodeARC black-box function task with the original descriptive function names left in place, which hint at what the hidden function does. Higher is better. Source: arxiv.org/abs/2503.23145.
Metric: Success Rate (%; descriptive function names, 1,114 functions). Source: arxiv.org. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 Mini | 59.5 | #283 |
| 2 | O1 Mini | 53.2 | #335 |
| 3 | DeepSeek R1 | 49.8 | #266 |
| 4 | Claude 3.7 Sonnet | 39 | #263 |
| 5 | Llama 3.1 405B Instruct | 38.6 | #472 |
| 6 | GPT-4o | 37.8 | #358 |
| 7 | DeepSeek V3 | 37.7 | #334 |
| 8 | Llama 3.1 70B Instruct | 30.1 | #570 |
| 9 | Qwen 2.5 72B Instruct | 30.1 | #470 |
| 10 | Qwen 2.5 7B Instruct | 29.2 | #866 |
| 11 | Llama 3.2 90B Vision Instruct | 28.4 | #626 |
| 12 | GPT-4o Mini | 26.1 | #601 |
| 13 | Mixtral 8x22B Instruct | 25.1 | #741 |
| 14 | Llama 3.2 11B Instruct | 24.9 | #1110 |
| 15 | Mixtral 8x7B Instruct | 20.3 | #1085 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #6 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=codearc-annotated · How It Works · Data refreshed daily, snapshot 2026-11-07.