VTC-Bench (Code Tools) - Math: leaderboard
Metric: Pass rate (%) on geometry math with auxiliary constructions (110 questions), answering VTC-Bench's 680 visual questions (538 multiple-choice, 142 open-ended) that call for chains of image operations with 32 OpenCV-based tools invoked by writing code (Qwen-Agent, or the Thyme framework for the open tool-use models), up to 20 rounds; rule matching plus a GPT-4o judge for open answers; proprietary models in high reasoning mode; higher is better. Source: arxiv.org. Saturation forecast: Around 2029. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.2 (High) | 45.45 | #105 (GPT-5.2) |
| 2 | Gemini 3 Pro (High) | 41.82 | #77 (Gemini 3 Pro) |
| 3 | O4 Mini (High) | 40 | #172 (O4 Mini) |
| 4 | Gemini 3 Flash (High) | 35.45 | #93 (Gemini 3 Flash) |
| 5 | Qwen 3 VL 235B A22B (Thinking) | 34.55 | #228 (Qwen 3 VL 235B A22B) |
| 6 | Qwen 3 VL 32B Instruct | 33.64 | #276 |
| 7 | Qwen 3 VL 32B (Thinking) | 31.82 | #287 (Qwen 3 VL 32B) |
| 8 | O3 (High) | 30.91 | #121 (O3) |
| 9 | Qwen 3 VL 235B A22B Instruct | 30 | #264 |
| 10 | Qwen 3 VL 30B A3B (Thinking) | 27.27 | #338 (Qwen 3 VL 30B A3B) |
| 11 | Qwen 3 VL 8B Instruct | 23.64 | #401 |
| 12 | Qwen 3 VL 30B A3B Instruct | 23.64 | #365 |
| 13 | Qwen 3 VL 8B (Thinking) | 22.73 | |
| 14 | GPT-4o | 20.91 | #333 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=vtc-bench-code-tools-math · How It Works · Data refreshed daily, snapshot 2026-10-11.