Hard Reasoning on HLE
37.7Pass@1Gemini-3.0 Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini-3.0 Pro2025.12 | 37.7 | |
| GPT-5 High2025.12 | 26.3 | |
| DeepSeek-V3.2thinking mode=true, template=step-by-step boxed2025.12 | 25.1 | |
| Kimi-K2thinking mode=true2025.12 | 23.9 | |
| DeepSeek-V3.2thinking mode=true, template=official2025.12 | 23.9 | |
| Claude-4.5-Sonnet2025.12 | 13.7 | |
| MiniMax M22025.12 | 12.5 | |
| Qwen3 + SFT + B-DAPO w/ recyclingSize=4B, k=2, Recycle=true2026.06 | 10.4 | |
| SmartSearch-3BSize=3B2026.06 | 9.8 | |
| Qwen3 + SFT + B-DAPO w/ recyclingSize=1.7B, k=2, Recycle=true2026.06 | 8 | |
| BehaviorPrime-1.7BSize=1.7B2026.06 | 7.8 | |
| ASearcher-Web-7BSize=7B2026.06 | 7 |