PRIME (Process-Outcome Alignment) - Math: leaderboard
Metric: Overall accuracy (%) of the model as a verifier on PRIME's 900 mathematics items (textbook and exam problems with a model-written solution, kept where GPT-OSS-120B's eight verification verdicts disagreed): the model must accept a solution only if both its derivation and its final answer are correct, scored against expert process-outcome labels; higher is better. Source: arxiv.org. Saturation forecast: Estimated already saturated. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (Thinking) | 86.67 | #91 (GPT-5) |
| 2 | Gemini 2.5 Pro | 86.11 | #145 |
| 3 | GPT-5.2 (Thinking) | 86 | #105 (GPT-5.2) |
| 4 | Grok 4 | 85.78 | #169 |
| 5 | DeepSeek V3.2 (Thinking) | 83.89 | #198 (DeepSeek V3.2) |
| 6 | Kimi K2 (Thinking) | 82.56 | #236 (Kimi K2) |
| 7 | GPT-5.2 Instant | 81.89 | #205 |
| 8 | Gemini 3 Pro | 81.67 | #77 |
| 9 | Claude Sonnet 4.5 (Thinking) | 81 | #138 (Claude Sonnet 4.5) |
| 10 | GLM-4.6 | 81 | #246 |
| 11 | Claude Sonnet 4 (Thinking) | 81 | #194 (Claude Sonnet 4) |
| 12 | Claude Opus 4 (Thinking) | 80.89 | #155 (Claude Opus 4) |
| 13 | Qwen 3 30B A3B | 79.67 | #488 |
| 14 | Claude Opus 4.5 (Thinking) | 79.33 | #79 (Claude Opus 4.5) |
| 15 | Qwen 3 4B 2507 (Thinking) | 76.22 | #525 (Qwen 3 4B 2507) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=prime-process-outcome-alignment-math · How It Works · Data refreshed daily, snapshot 2026-10-11.