ArXivMath 04/2026
Math · 2026-05-01
This MathArena track evaluates models on the ArXivMath 04/2026 problem set. Published scores average correctness across observed problem attempts; the item count records distinct problems, not a single-run binomial denominator.
Top models (higher is better)
| Model | Score |
|---|---|
| GPT-6 Astra | 90.8 |
| Claude Fable 5.1 | 77.5 |
| Claude Fable 5 | 72.5 |
| GPT-5.5 | 68.8 |
| Muse Spark 1.3 | 66.7 |
| Gemini 3.8 Flash | 64.2 |
| Gemini 3.1 Pro Preview | 63.8 |
| Opus 4.8 | 62.5 |
| Opus 4.7 | 57.5 |
| Gemini 3.7 Flash | 57.5 |
| Kimi K3 | 54.2 |
| Gemini 3.5 Flash | 52.5 |
| Gemini 3.6 Flash | 52.5 |
| DeepSeek-V4-Pro | 51.3 |
| DeepSeek-V4-Flash | 47.5 |
| GLM-5.2 | 47.5 |
| DeepSeek V4.1 Flash | 45.0 |
| GLM-5.1 | 36.3 |
| Step 3.7 Flash | 35.8 |
| Qwen3.6 35B A3B | 33.8 |
| Step 3.5 Flash | 33.8 |
| Qwen3.5-2B | 1.9 |