LLM Stats (MM-MT-Bench): leaderboard
Multimodal multi-turn benchmark evaluating ability to maintain context and reason across multiple conversational turns with interleaved images and text.
Metric: Score (%). Source: llm-stats.com. Saturation forecast: Around 2032. 17 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 235B A22B Instruct | 8.5 | #273 |
| 2 | Qwen 3 VL 235B A22B (Thinking) | 8.5 | #238 (Qwen 3 VL 235B A22B) |
| 3 | Ministral-3-14B-Instruct-2512 | 8.49 | #615 |
| 4 | Qwen 3 VL 32B Instruct | 8.4 | #283 |
| 5 | Qwen 3 VL 32B (Thinking) | 8.3 | #294 (Qwen 3 VL 32B) |
| 6 | Qwen 3 VL 30B A3B Instruct | 8.1 | #384 |
| 7 | Ministral-3-8B-Instruct-2512 | 8.08 | #661 |
| 8 | Qwen 3 VL 8B (Thinking) | 8 | |
| 9 | Qwen 3 VL 30B A3B (Thinking) | 7.9 | #351 (Qwen 3 VL 30B A3B) |
| 10 | Ministral-3-3B-Instruct-2512 | 7.83 | #855 |
| 11 | Qwen 3 VL 8B Instruct | 7.7 | #420 |
| 12 | Qwen 3 VL 4B (Thinking) | 7.7 | #480 (Qwen 3 VL 4B) |
| 13 | Qwen 3 VL 4B Instruct | 7.5 | #515 |
| 14 | Qwen2.5-Omni-7B | 6 | #624 |
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-mm-mt-bench · How It Works · Data refreshed daily, snapshot 2026-10-20.