MathVision (MATH-Vision) — Explicit tool-use axis; scores far higher (84.6-96.1) consistent with tool assistance.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | GPT-5.4 | OpenAI | 96.1 | 1 | 2026-05-03 |
| 2 | Gemini 3.1 Pro | 95.7 | 2 | 2026-05-03 | |
| 3 | Kimi K2.6 | Moonshot | 93.2 | 1 | 2026-05-03 |
| 4 | Kimi K2.5 | Moonshot | 85 | 1 | 2026-05-03 |
| 5 | Claude Opus 4.6 | Anthropic | 84.6 | 1 | 2026-05-03 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.