Polaris-Bench: leaderboard
As current Multimodal Large Language Models rapidly saturate canonical visual reasoning benchmarks, a key question emerges: do these strong scores genuinely reflect robust visual understanding? We identify a pervasive vulnerability, the Cartesian Shortcut: visual reasoning benchmarks prevalently build on orthogonal grid-based layouts that can be readily discretized into explicit textual coordinate
Metric: Overall C (self-reported). Source: benchmarklist.com. Saturation forecast: Around June 2027. 15 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 82.6 | #58 |
| 2 | GPT-5.2 | 77.4 | #110 |
| 3 | Qwen 3.5 397B A17B | 72.9 | #146 |
| 4 | Gemini 3 Flash | 71 | #99 |
| 5 | Kimi K2.5 | 69 | #144 |
| 6 | Gemma 4 31B | 60.5 | #200 |
| 7 | Gemma 4 26B | 47.2 | #292 |
| 8 | Claude Sonnet 4.6 | 44.4 | #90 |
| 9 | Gemini 2.5 Pro | 38.4 | #152 |
| 10 | Grok 4 | 33 | #181 |
| 11 | Gemini 2.5 Flash | 32.2 | #245 |
| 12 | Grok 4 Fast (Reasoning) | 31 | #254 (Grok 4 Fast) |
| 13 | Mistral Small 3.1 | 19.4 | #610 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=polaris-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.