Multimodal Reasoning on MathVista
89.8Pass@1Gemini 3-Pro
Evaluation Results
| Method | Links | |
|---|---|---|
| Gemini 3-Pro2026.03 | 89.8 | |
| Seed1.82026.03 | 87.7 | |
| Seed 1.5-VLthinking=true, decoding=greedy2025.05 | 85.6 | |
| Seed1.5-VL Thinking2026.03 | 85.6 | |
| Seed 1.5-VLthinking=false, decoding=greedy2025.05 | 83 | |
| Gemini 1.5 Prothinking=true, decoding=greedy2025.05 | 82.7 | |
| Gemini 2.5-Prosourced from technical report=true2026.03 | 82.7 | |
| GPT-5.1 High2026.03 | 80.6 | |
| Claude Sonnet-4.52026.03 | 80.4 | |
| Qwen 2.5-VL 72Bthinking=false, decoding=greedy2025.05 | 74.8 | |
| Claude 3.7 Sonnetthinking=true, decoding=sampling2025.05 | 74.5 | |
| RAPBackbone=Qwen2.5-VL-7b, Sample=5,159, Time (h)=52.82025.06 | 73.2 | |
| OpenAI o1thinking=true, decoding=greedy2025.05 | 71.8 | |
| LIMRBackbone=Qwen2.5-VL-7b, Sample=8,136, Time (h)=1222025.06 | 71.1 | |
| Qwen2.5-VL-7b-FullBackbone=Qwen2.5-VL-7b, Sample=54,931, Time (h)=93.22025.06 | 70.7 | |
| LIMOBackbone=Qwen2.5-VL-7b, Sample=4,093, Time (h)=111.92025.06 | 69.9 | |
| Qwen2.5-VL-7bBackbone=Qwen2.5-VL-7b2025.06 | 68.7 | |
| s1Backbone=Qwen2.5-VL-7b, Sample=1,000, Time (h)=55.92025.06 | 68.5 | |
| RAPBackbone=Qwen2.5-VL-3b, Sample=4,374, Time (h)=322025.06 | 64.9 | |
| Qwen2.5-VL-3b-FullBackbone=Qwen2.5-VL-3b, Sample=54,931, Time (h)=46.52025.06 | 64.5 | |
| GPT-4othinking=false, decoding=greedy2025.05 | 63.8 | |
| Socratic-Solver-Geo (Stage3)Data Scale=2.5k, Curriculum Stage=32026.02 | 63.55 | |
| GeoReasoningData Scale=10k2026.02 | 63.4 | |
| LIMRBackbone=Qwen2.5-VL-3b, Sample=21,303, Time (h)=60.92025.06 | 63.1 | |
| R-CoTData Scale=7.2k2026.02 | 62.6 | |
| s1Backbone=Qwen2.5-VL-3b, Sample=1,000, Time (h)=38.42025.06 | 62.6 | |
| KD (Our Synthesis)Data Scale=2.5k2026.02 | 62.4 | |
| Socratic-Solver-Geo (Stage2)Data Scale=1k, Curriculum Stage=22026.02 | 62.3 | |
| KD (Geo3K)Data Scale=3k2026.02 | 62.1 | |
| Geo170k (G-LLaVA)Data Scale=10k2026.02 | 62 | |
| TrustGeoGenData Scale=10k2026.02 | 61.8 | |
| LIMOBackbone=Qwen2.5-VL-3b, Sample=2,679, Time (h)=120.52025.06 | 61.8 | |
| PGPS9KData Scale=10k2026.02 | 61.3 | |
| Qwen2.5-VL-3bBackbone=Qwen2.5-VL-3b2025.06 | 61.3 | |
| Socratic-Solver-Geo (Stage1)Data Scale=0.4k, Curriculum Stage=12026.02 | 61.2 | |
| Qwen2.5-VL-7B-InstructMode=Zero-shot2026.02 | 61.1 |