FlagEval VQA - Multi-Image Analysis: leaderboard
BAAI FlagEval’s multimodal VQA leaderboard, Multi-Image Analysis slice: questions answerable only by comparing several images.
Metric: Score. Source: flageval.baai.ac.cn. Saturation forecast: Around November 2026. 24 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 (High) | 62.5 | #98 (GPT-5) |
| 2 | O3 Mini (Low) | 61.7 | #272 (O3 Mini) |
| 3 | O3 Mini (High) | 60 | #272 (O3 Mini) |
| 4 | GPT-5 (Low) | 58.3 | #98 (GPT-5) |
| 5 | GPT-4.1 | 57.5 | #248 |
| 6 | Gemini 2.5 Pro | 51.7 | #152 |
| 7 | O4 Mini (High) | 51.7 | #180 (O4 Mini) |
| 8 | Gemini 2.5 Flash (Non-reasoning) | 46.7 | #245 (Gemini 2.5 Flash) |
| 9 | GLM-4.5V | 45 | #346 |
| 10 | Gemini 2.5 Flash (Thinking) | 44.2 | #245 (Gemini 2.5 Flash) |
| 11 | O4 Mini (Low) | 43.3 | #180 (O4 Mini) |
| 12 | Step-3 | 42.5 | #365 |
| 13 | ERNIE-4.5-VL-424B-A47B | 36.7 | #391 |
| 14 | Claude Sonnet 4 (Thinking) | 33.3 | #199 (Claude Sonnet 4) |
| 15 | Claude Sonnet 4 (Non-reasoning) | 29.2 | #199 (Claude Sonnet 4) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=flageval-vqa-multi-image-analysis · How It Works · Data refreshed daily, snapshot 2026-10-20.