Token Rewards Evaluation - AIME25: leaderboard
Token Rewards Evaluation - AIME25. Reported benchmark pass@1;Qwen3 modes explicit. GRPO/P2T trained variants retain independent identities; source Table1. Metric: Pass@1 (%). Primary published table; source configurations remain distinct.
Metric: Pass@1 (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 1.7B (Thinking) | 35.4 | #1178 (Qwen 3 1.7B) |
| 2 | DeepSeek R1 Distill Qwen 1.5B | 22.8 | #1535 |
| 3 | Qwen2.5-Math-7B | 9.8 | #1467 |
| 4 | Qwen 3 1.7B (Non-reasoning) | 8.8 | #1178 (Qwen 3 1.7B) |
| 5 | Llama 3.2 3B Instruct | 0.2 | #1336 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=token-rewards-evaluation-aime25 · How It Works · Data refreshed daily, snapshot 2026-10-16.