HELM Reasoning - Math500: leaderboard
The "Math500" subtask of HELM Reasoning. Stanford HELM's reasoning project: competition mathematics, code and logical-reasoning scenarios (AIME, MATH, GPQA-style sets) run through the HELM harness with accuracy metrics.
Metric: Acc. Source: crfm.stanford.edu. Saturation forecast: Estimated already saturated. 11 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 Mini (2025-08-07) | 77.2 | #174 |
| 2 | Qwen 3 8B | 76.4 | #678 |
| 3 | Qwen 3 4B | 75.2 | #840 |
| 4 | Qwen 3 1.7B | 71.4 | #1200 |
| 5 | Qwen3-8B-Base | 61.6 | #628 |
| 6 | Qwen 3 0.6B | 57.2 | #1464 |
| 7 | Qwen3-4B-Base | 56.4 | #901 |
| 8 | Qwen3-1.7B-Base | 47.6 | #1189 |
| 9 | Qwen3-0.6B-Base | 37 | #1493 |
No result here: #4 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=helm-reasoning-math500 · How It Works · Data refreshed daily, snapshot 2026-10-23.