MathVision
We show this table for reference; we do not rank on it.
A visual mathematics benchmark that tests whether a model can solve math problems grounded in diagrams, equations, figures, and other visual inputs.
Benchmark score on MathVision — November 13, 2026
We compile the MathVision rows from provider self-reports and secondary reports. Qwen3.8 Max leads the table at 95.2%, followed by Kimi K3 (94.3%) and Inkling 2 Small (92.8%). We do not use these results to rank models overall.
Qwen3.8 Max
Alibaba
Reported by AlibabaKimi K3
Moonshot AI
Reported by Moonshot AIInkling 2 Small
Thinking Machines Lab
Reported by Thinking Machines Lab24 modelsMultimodal & GroundedCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (24 models)
None of the 24 results on this page has a documented run by the benchmark or an independent evaluator: 21 reported by the labs that built the models and 3 reported in another publisher’s document.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Qwen3.8 MaxAlibabaReported by Alibaba | 95.2% | Reported by Alibaba | Not reported | Open |
| 2 | Kimi K3Moonshot AIReported by Moonshot AI | 94.3% | Reported by Moonshot AI | Not reported | Open |
| 3 | Inkling 2 SmallThinking Machines LabReported by Thinking Machines Lab | 92.8% | Reported by Thinking Machines Lab | Not reported | Closed |
| 4 | Seed 2.1 ProByteDanceReported by ByteDance | 92.6% | Reported by ByteDance | Not reported | Closed |
| 5 | Qwen3.8-Omni-FlashAlibabaReported by Alibaba | 91.8% | Reported by Alibaba | Not reported | Closed |
| 6 | Qwen3.8-Flash-NextAlibabaReported by Alibaba | 90.6% | Reported by Alibaba | Not reported | Open |
| 7 | Qwen3.7 PlusAlibabaReported by Alibaba | 90.3% | Reported by Alibaba | Not reported | Closed |
| 8 | Seed 2.1 TurboByteDanceReported by ByteDance | 90.1% | Reported by ByteDance | Not reported | Closed |
| 9 | Qwen3.8-27BAlibabaReported by Alibaba | 90.0% | Reported by Alibaba | Not reported | Open |
| 10 | Qwen3.5 397BAlibabaReported by Alibaba | 88.6% | Reported by Alibaba | Not reported | Open |
| 11 | Qwen3.6 PlusAlibabaReported by Alibaba | 88.0% | Reported by Alibaba | Not reported | Closed |
| 12 | dots3-note PreviewDots StudioReported by Dots Studio | 87.7% | Reported by Dots Studio | Not reported | Open |
| 13 | Kimi K2.6Moonshot AIReported by Moonshot AI | 87.4% | Reported by Moonshot AI | Not reported | Open |
| 14 | Gemini 3 ProGoogleAnother lab’s comparison table | 86.6% | Another lab’s comparison table | Not reported | Closed |
| 15 | Qwen3.5-122B-A10BAlibabaReported by Alibaba | 86.2% | Reported by Alibaba | Not reported | Open |
| 16 | Qwen3.5-27BAlibabaReported by Alibaba | 86.0% | Reported by Alibaba | Not reported | Open |
| 17 | Kimi K2.5Moonshot AIReported by Moonshot AI | 84.2% | Reported by Moonshot AI | Not reported | Open |
| 18 | Qwen3.5-35B-A3BAlibabaReported by Alibaba | 83.9% | Reported by Alibaba | Not reported | Open |
| 19 | GPT-5.2OpenAIAnother lab’s comparison table | 83.0% | Another lab’s comparison table | Not reported | Closed |
| 20 | Gemma 4 12BGoogleReported by Google | 79.7% | Reported by Google | Not reported | Open |
| 21 | Nemotron 3.5 Super VL 120B A12BNVIDIAReported by NVIDIA | 78.2% | Reported by NVIDIA | Not reported | Open |
| 22 | Nemotron 3.5 Super VL 120B A12B NVFP4NVIDIAReported by NVIDIA | 77.8% | Reported by NVIDIA | Not reported | Open |
| 23 | Claude Opus 4.5AnthropicAnother lab’s comparison table | 74.3% | Another lab’s comparison table | Not reported | Closed |
| 24 | LFM3-VL-2BLiquidAIReported by LiquidAI | 51.6% | Reported by LiquidAI | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
Among the reported MathVision rows, Qwen3.8 Max is first at 95.2%. The third row is 2.4 points behind. The broader top-10 range is 6.6 points, so many of the published results sit in a relatively narrow band.
24 models have been evaluated on MathVision. The benchmark falls in the Multimodal & Grounded category. MathVision is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About MathVision
Year
2026
Tasks
Visually grounded math problems
Format
Image + math reasoning
Difficulty
Advanced multimodal mathematics
MathVision matters because text-only math ability does not guarantee strong performance when the relevant information is embedded in images, geometry diagrams, or formatted equations.
Freshness and provenance
Version
MathVision 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does MathVision measure?
A visual mathematics benchmark that tests whether a model can solve math problems grounded in diagrams, equations, figures, and other visual inputs.
Which model scores highest on MathVision?
Qwen3.8 Max by Alibaba currently leads with a score of 95.2% on MathVision.
How many models are evaluated on MathVision?
24 AI models have published results on MathVision in the BenchLM catalog.
Compare top models on MathVision
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.