RealWorldQA — Real-world visual QA benchmark, accuracy metric.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Qwen2.5 VL 72B | Alibaba | 75.7 | 1 | 2026-06-05 |
| 2 | GPT-4o | OpenAI | 75.4 | 2 | 2026-06-05 |
| 3 | Kimi VL A3B Thinking | Moonshot | 70 | 2 | 2026-06-12 |
| 4 | Kimi VL A3B Instruct | Moonshot | 68.1 | 2 | 2026-06-12 |
| 5 | GPT-4o mini | OpenAI | 67.1 | 1 | 2026-06-05 |
| 6 | Gemma 3 27B | 62.5 | 1 | 2026-06-05 | |
| 7 | Gemma 3 12B | 59.1 | 2 | 2026-06-05 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.