GeoAux-Bench (Core): leaderboard
Metric: Accuracy (%) on all 2,379 questions of GeoAux-Bench-Core (expert-annotated geometry problems needing auxiliary constructions; final answers checked by a Qwen3-30B-A3B-Thinking-2507 verifier; greedy decoding, 8,192 new tokens); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 21 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 2.5 Pro | 83.16 | #145 |
| 2 | GPT-5 | 80.62 | #91 |
| 3 | Gemini 2.5 Flash | 79.53 | #237 |
| 4 | Qwen 3 VL 235B A22B Instruct | 74.85 | #264 |
| 5 | Qwen 3 VL 30B A3B Instruct | 70.25 | #365 |
| 6 | Doubao-Seed-1.6 (Thinking) | 65 | |
| 7 | Qwen 3 VL 235B A22B (Thinking) | 62.25 | #228 (Qwen 3 VL 235B A22B) |
| 8 | Qwen 3 VL 30B A3B (Thinking) | 58.75 | #338 (Qwen 3 VL 30B A3B) |
| 9 | Claude Opus 4.1 (Thinking) | 55.82 | #132 (Claude Opus 4.1) |
| 10 | Claude Sonnet 4.5 (Thinking) | 55.03 | #138 (Claude Sonnet 4.5) |
| 11 | GLM-4.5V (Non-reasoning) | 40.24 | #339 (GLM-4.5V) |
| 12 | InternVL3-8B | 35.17 | #606 |
| 13 | GLM-4.1V-9B (Thinking) | 31.76 | #457 (GLM-4.1V-9B) |
| 14 | GPT-4o | 28.13 | #333 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=geoaux-bench-core · How It Works · Data refreshed daily, snapshot 2026-10-11.