Smolagents - MATH (Code Agent): leaderboard
Metric: Accuracy (%). Source: huggingface.co. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | O3 Mini | 98 |
| 2 | Claude 3.7 Sonnet (20250219) | 96 |
| 3 | O1 | 96 |
| 4 | DeepSeek R1 | 94 |
| 5 | QwQ-32B | 94 |
| 6 | DeepSeek R1 Distill Qwen 32B | 92 |
| 7 | GPT-4.5 (Preview) | 92 |
| 8 | Claude 3.5 Sonnet | 86 |
| 9 | Qwen 2.5 Coder 32B Instruct | 84 |
| 10 | GPT-4o | 78 |
| 11 | Qwen 2.5 72B Instruct | 76 |
| 12 | Llama 3.3 70B Instruct | 72 |
| 13 | Llama 3.1 70B Instruct | 70 |
| 14 | Llama 3.2 3B Instruct | 40 |
| 15 | Llama 3.1 8B Instruct | 14 |
Interactive version: theaggregate.ai/benchmark?slug=smolagents-math-code-agent · How It Works · Data refreshed daily, snapshot 2026-10-09.