LLM Stats (BLINK): leaderboard
14 visual perception tasks humans solve in a 'blink' but multimodal LLMs struggle with, testing core visual abilities like depth estimation and spatial reasoning.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Seed 2.1 Pro | 81.4 | #67 |
| 2 | Seed 2.1 Turbo | 79.4 | #86 |
| 3 | Seed 1.8 | 74.3 | #139 |
| 4 | Seed 2.0 Mini | 73.4 | #276 |
| 5 | Qwen 3 VL 235B A22B Instruct | 70.7 | #273 |
| 6 | Qwen 3 VL 8B Instruct | 69.1 | #420 |
| 7 | Qwen 3 VL 8B (Thinking) | 68.7 | |
| 8 | Qwen 3 VL 32B (Thinking) | 68.5 | #294 (Qwen 3 VL 32B) |
| 9 | Qwen 3 VL 30B A3B Instruct | 67.7 | #384 |
| 10 | Qwen 3 VL 32B Instruct | 67.3 | #283 |
| 11 | Qwen 3 VL 235B A22B (Thinking) | 67.1 | #238 (Qwen 3 VL 235B A22B) |
| 12 | Qwen 3 VL 4B Instruct | 65.8 | #515 |
| 13 | Qwen 3 VL 30B A3B (Thinking) | 65.4 | #351 (Qwen 3 VL 30B A3B) |
| 14 | Qwen 3 VL 4B (Thinking) | 63.4 | #480 (Qwen 3 VL 4B) |
| 15 | Phi-4 Multimodal Instruct | 61.3 | #906 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=llm-stats-blink · How It Works · Data refreshed daily, snapshot 2026-10-20.