GR-Ben: leaderboard
Currently, process reward models (PRMs) have exhibited remarkable potential for test-time scaling. Since large language models (LLMs) regularly generate flawed intermediate reasoning steps when tackling a broad spectrum of reasoning and decision-making tasks, PRMs are required to possess capabilities for detecting process-level errors in real-world scenarios. However, existing benchmarks primarily
Metric: Average (self-reported). Source: benchmarklist.com. Saturation forecast: Around February 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash | 60.5 | #99 |
| 2 | DeepSeek V3.2 | 55.4 | #205 |
| 3 | Kimi K2 0905 | 49.1 | #288 |
| 4 | GPT-5.2 | 47 | #110 |
| 5 | Qwen 3 32B | 36.7 | #443 |
| 6 | Gemma 3 27B (IT) | 34.7 | #523 |
| 7 | Qwen 3 8B | 34.5 | #674 |
| 8 | Qwen 3 14B | 34.4 | #537 |
| 9 | Qwen 3 4B | 29.3 | #829 |
| 10 | Gemma 3 12B (IT) | 25.2 | #664 |
| 11 | Llama 3.3 70B Instruct | 22.3 | #534 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=gr-ben · How It Works · Data refreshed daily, snapshot 2026-10-20.