LetTokenBurnMathVis: Can AI Reason through Drawing? A Benchmark for Mathematical Figure Generation
A benchmark of 512 tasks evaluating LLMs' ability to generate executable TikZ/SVG code for advanced mathematical figures from statements.
$ npm i -g lettokenburn-cli$ ltb bench info mathvis
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| GPT-6 Astra | 31.3 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Claude Fable 5 | 26.6 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Gemini 3.1 Pro | 22.9 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| GPT-5.6 Sol | 21.5 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Claude Opus 4.8 | 20.9 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Doubao-Seed-2.1-Pro | 20.9 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| GLM-5.2 | 18.9 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Doubao Seed 2.0 Pro | 18.2 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| DeepSeek V4 Flash 0731 | 17.2 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Claude Sonnet 4.6 | 16.8 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Doubao Seed 2.1 Turbo | 16.2 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Gemini-3.5-flash | 16.0 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
MathVis comprises 512 tasks covering 27 advanced mathematical sub-disciplines. Each task includes a statement, reference diagram, and gold intermediate representation to test spatial alignment and code generation.
Evaluation uses atomic rubrics and error codes to separate programming from geometric errors. Top models achieve low Pass@1 scores (~31%), indicating significant difficulty in visual reasoning.