Mathematical Reasoning on MathVerse mini (test)
66.9AccuracyHuman
Evaluation results
| Method | Links | |
|---|---|---|
| HumanSize=–2026.03 | 66.9 | |
| Qwen2.5-VL-72B InstructZero-shot=true2026.02 | 57.2 | |
| Qwen2.5-VL-32B + AT-RL (Ours)Zero-shot=true2026.02 | 56.6 | |
| Gemini 2.0 FlashZero-shot=true2026.02 | 54.6 | |
| Qwen2.5-VL-32B + VPPOZero-shot=true2026.02 | 54.5 | |
| GPT-4oSize=–2026.03 | 50.5 | |
| Qwen2.5-VL-32B InstructZero-shot=true2026.02 | 49.3 | |
| Claude 3.5 SonnetZero-shot=true2026.02 | 47.8 | |
| LatentGeoSize=7B2026.03 | 41.4 | |
| OpenAI GPT-4oZero-shot=true2026.02 | 41.2 | |
| InternVL2.5-8BSize=8B2026.03 | 37.8 | |
| Gemini-1.5-ProSize=–2026.03 | 35.5 | |
| MAVIS-7BSize=7B2026.03 | 34.7 | |
| LLaVA-NeXT-34BSize=34B2026.03 | 33.9 | |
| InfiMM-MathSize=7B2026.03 | 33.3 | |
| Math-PUMA-Qwen2-7BSize=7B2026.03 | 33.3 | |
| InternVL2-8BSize=8B2026.03 | 32.5 | |
| Math-PUMA-DeepSeek-MathSize=7B2026.03 | 30.8 | |
| MultiMath-7BSize=7B2026.03 | 26 | |
| Math-LLaVA-13BSize=13B2026.03 | 22.4 | |
| Qwen-VL-PlusSize=–2026.03 | 21.4 | |
| DeepSeek-VLSize=7B2026.03 | 18.4 | |
| SPHINX-V2-13BSize=13B2026.03 | 17.3 | |
| LLaVA-1.5-13BSize=13B2026.03 | 12.9 | |
| RandomSize=–2026.03 | 12.4 | |
| mPLUG-Owl2-7BSize=7B2026.03 | 10 |