We-Math
We show this table for reference; we do not rank on it.
A multimodal math benchmark for visually grounded mathematical reasoning and answer generation.
Benchmark score on We-Math — November 13, 2026
We compile the We-Math rows from provider self-reports and secondary reports. Qwen3.5 397B leads the table at 87.9%, followed by Gemini 3 Pro (86.9%) and Kimi K2.5 (84.7%). We do not use these results to rank models overall.
Qwen3.5 397B
Alibaba
Reported by AlibabaGemini 3 Pro
Kimi K2.5
Moonshot AI
Another lab’s comparison table6 modelsMultimodal & GroundedCurrentDisplay onlyUpdated November 13, 2026
Benchmark score table (6 models)
None of the 6 results on this page has a documented run by the benchmark or an independent evaluator: 2 reported by the labs that built the models and 4 reported in another publisher’s document.
| Rank | Model / configuration | Score | Who ran it | Parameters (B) | Open / closed |
|---|---|---|---|---|---|
| 1 | Qwen3.5 397BAlibabaReported by Alibaba | 87.9% | Reported by Alibaba | Not reported | Open |
| 2 | Gemini 3 ProGoogleAnother lab’s comparison table | 86.9% | Another lab’s comparison table | Not reported | Closed |
| 3 | Kimi K2.5Moonshot AIAnother lab’s comparison table | 84.7% | Another lab’s comparison table | Not reported | Open |
| 4 | GPT-5.2OpenAIAnother lab’s comparison table | 79.0% | Another lab’s comparison table | Not reported | Closed |
| 5 | Claude Opus 4.5AnthropicAnother lab’s comparison table | 70.0% | Another lab’s comparison table | Not reported | Closed |
| 6 | LFM3-VL-2BLiquidAIReported by LiquidAI | 68.6% | Reported by LiquidAI | Not reported | Open |
- Lab-reported
- Benchmark-run
- Independent
Among the reported We-Math rows, Qwen3.5 397B is first at 87.9%. The third row is 3.2 points behind. The broader top-10 range is 19.3 points, so the table still separates the published systems.
6 models have been evaluated on We-Math. The benchmark falls in the Multimodal & Grounded category. We-Math is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About We-Math
Year
2026
Tasks
Visually grounded math problems
Format
Multimodal mathematical reasoning
Difficulty
Advanced multimodal mathematics
We-Math is useful as a visual-math stress test because it combines symbolic reasoning with figure understanding. It helps reveal whether a model's math strength transfers into multimodal settings.
Freshness and provenance
Version
We-Math 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 We-Math measure?
A multimodal math benchmark for visually grounded mathematical reasoning and answer generation.
Which model scores highest on We-Math?
Qwen3.5 397B by Alibaba currently leads with a score of 87.9% on We-Math.
How many models are evaluated on We-Math?
6 AI models have published results on We-Math in the BenchLM catalog.
Compare top models on We-Math
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.