CFE-Bench (Text): leaderboard
Metric: Question accuracy (%): share of the 305 text-only CFE-Bench problems (university STEM homework and exam problems with instructor-verified solutions) for which every annotated answer variable is correct; answers extracted and verified per annotated target variable by a GPT-5-mini judge (gpt-5-mini-2025-08-07) against expert-annotated values; chain-of-thought prompting, thinking capped at 16,000 tokens; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 64.92 | #54 |
| 2 | Gemini 3 Flash (Preview) | 58.69 | #78 |
| 3 | Gemini 3 Pro (Preview) | 58.03 | #64 |
| 4 | Claude Opus 4.6 | 52.79 | #60 |
| 5 | GPT-5.2 | 51.15 | #105 |
| 6 | Qwen 3.5 397B A17B | 48.52 | #141 |
| 7 | Qwen 3.5 Plus | 48.2 | #123 |
| 8 | Grok 4 | 47.54 | #169 |
| 9 | Kimi K2.5 | 43.93 | #139 |
| 10 | Grok 4.1 Fast (Reasoning) | 43.61 | #208 (Grok 4.1 Fast) |
| 11 | DeepSeek V3.2 (Thinking) | 43.28 | #198 (DeepSeek V3.2) |
| 12 | Claude Opus 4.5 | 41.97 | #79 |
| 13 | GLM-5 | 41.64 | #137 |
| 14 | DeepSeek V3.2 (Non-reasoning) | 41.64 | #198 (DeepSeek V3.2) |
| 15 | Kimi K2 (Thinking) | 39.02 | #236 (Kimi K2) |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=cfe-bench-text · How It Works · Data refreshed daily, snapshot 2026-10-11.