MLVU (Multi-task Long Video Understanding) — Explicit MCQ subset + accuracy metric axis, distinct from plain MLVU.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Kimi VL A3B Instruct | Moonshot | 74.2 | 1 | 2026-06-12 |
| 2 | GPT-4o | OpenAI | 64.6 | 1 | 2026-06-05 |
| 3 | Gemma 3 12B | 52.3 | 1 | 2026-06-05 | |
| 4 | GPT-4o mini | OpenAI | 48.1 | 1 | 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.