LLM Stats (MMStar): leaderboard
1,500 carefully curated multimodal questions requiring genuine visual reasoning, filtered to remove items solvable by text-only models or random guessing.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around January 2027. 27 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3.6 Plus | 83.3 | #120 |
| 2 | Qwen 3.5 122B A10B | 82.9 | #173 |
| 3 | Qwen 3.5 35B A3B | 81.9 | #256 |
| 4 | Qwen 3.6 27B | 81.4 | #170 |
| 5 | Qwen 3.5 27B | 81 | #187 |
| 6 | Seed 1.8 | 79.9 | #139 |
| 7 | Qwen 3 VL 32B (Thinking) | 79.4 | #294 (Qwen 3 VL 32B) |
| 8 | Seed 2.0 Mini | 79.1 | #276 |
| 9 | Qwen 3 VL 235B A22B (Thinking) | 78.7 | #238 (Qwen 3 VL 235B A22B) |
| 10 | Qwen 3 VL 235B A22B Instruct | 78.4 | #273 |
| 11 | Qwen 3 VL 32B Instruct | 77.7 | #283 |
| 12 | Qwen 3 VL 30B A3B (Thinking) | 75.5 | #351 (Qwen 3 VL 30B A3B) |
| 13 | Qwen 3 VL 8B (Thinking) | 75.3 | |
| 14 | EXAONE 4.5 33B | 74.9 | #396 |
| 15 | Qwen 3 VL 4B (Thinking) | 73.2 | #480 (Qwen 3 VL 4B) |
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-mmstar · How It Works · Data refreshed daily, snapshot 2026-10-20.