Multimodal Mathematical Reasoning on MathVista 14 (1000)
90.2Macro ScoreQwen3.5-397B-A17B*
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Qwen3.5-397B-A17B*Model Type=Open-source LMM (Large Scale > 30B), Evaluation Protocol=API calls2026.05 | 90.2 | — | — | |
| Gemini-3-Pro*Model Type=Closed-source LMM, Evaluation Protocol=API calls2026.05 | 88.5 | — | — | |
| Doubao-Seed-1.8*Model Type=Closed-source LMM, Evaluation Protocol=API calls2026.05 | 86.3 | — | — | |
| Qwen3-VL-235B-A22B*Model Type=Open-source LMM (Large Scale > 30B), Evaluation Protocol=API calls2026.05 | 84.9 | — | — | |
| GPT-5*Model Type=Closed-source LMM, Evaluation Protocol=API calls2026.05 | 81.9 | — | — | |
| Qwen3-VL-30B-A3B*Model Type=Open-source LMM (Large Scale > 30B), Evaluation Protocol=API calls2026.05 | 81.2 | — | — | |
| Qwen3VL-8B-instruct + GeoSym EntryBase Model=Qwen3-VL, Parameter Scale=8B, Method Component=GeoSym Entry2026.05 | 76.6 | 92.31 | 90.38 | |
| Qwen3VL-8B-instruct + GeoSym HardBase Model=Qwen3-VL, Parameter Scale=8B, Method Component=GeoSym Hard2026.05 | 76.6 | 92.79 | 90.8 | |
| Qwen3VL-8B-instructBase Model=Qwen3-VL, Parameter Scale=8B2026.05 | 75.8 | 87.5 | 85.77 | |
| Qwen3VL-4B-Instruct + GeoSym HardBase Model=Qwen3-VL, Parameter Scale=4B, Method Component=GeoSym Hard2026.05 | 74.8 | 88.94 | 86.19 | |
| Qwen3VL-4B-InstructBase Model=Qwen3-VL, Parameter Scale=4B2026.05 | 73.4 | 83.65 | 82.43 | |
| Qwen3VL-4B-Instruct + GeoSym EntryBase Model=Qwen3-VL, Parameter Scale=4B, Method Component=GeoSym Entry2026.05 | 72.8 | 87.02 | 84.52 | |
| Qwen2.5VL-7B-Instruct + GeoSym HardBase Model=Qwen2.5-VL, Parameter Scale=7B, Method Component=GeoSym Hard2026.05 | 69.4 | 70.67 | 68.62 | |
| Qwen2.5VL-7B-Instruct + GeoSym EntryBase Model=Qwen2.5-VL, Parameter Scale=7B, Method Component=GeoSym Entry2026.05 | 68.5 | 67.79 | 66.53 | |
| Qwen2.5VL-7B-InstructBase Model=Qwen2.5-VL, Parameter Scale=7B2026.05 | 67.9 | 70.19 | 69.87 | |
| Qwen3VL-8B-instruct + GeoTrust-trainBase Model=Qwen3-VL, Parameter Scale=8B, External Synthesis Method=GeoTrust-train2026.05 | 63.2 | 78.85 | 75.73 | |
| Qwen2.5VL-7B-Instruct + GeoTrust-trainBase Model=Qwen2.5-VL, Parameter Scale=7B, External Synthesis Method=GeoTrust-train2026.05 | 63.1 | 65.38 | 64.02 | |
| Qwen3VL-8B-instruct + TR-GeoMMBase Model=Qwen3-VL, Parameter Scale=8B, External Synthesis Method=TR-GeoMM2026.05 | 61.8 | 64.9 | 64.85 | |
| Qwen2.5VL-7B-Instruct + TR-GeoMMBase Model=Qwen2.5-VL, Parameter Scale=7B, External Synthesis Method=TR-GeoMM2026.05 | 61.7 | 53.85 | 54.81 | |
| Qwen2.5VL-3B-Instruct + GeoSym EntryBase Model=Qwen2.5-VL, Parameter Scale=3B, Method Component=GeoSym Entry2026.05 | 60.1 | 53.84 | 54.18 | |
| Qwen2.5VL-3B-InstructBase Model=Qwen2.5-VL, Parameter Scale=3B2026.05 | 59.6 | 50.48 | 50.63 | |
| Qwen2.5VL-3B-Instruct + GeoSym HardBase Model=Qwen2.5-VL, Parameter Scale=3B, Method Component=GeoSym Hard2026.05 | 52.1 | 57.21 | 57.74 |