GAIA — no-file-attachment restriction — GAIA excluding file-attachment tasks — distinct from text-only (358).
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Step 3.5 Flash | StepFun | 84.5 | 3 | 2026-05-30 |
| 2 | MiMo V2 Flash | Xiaomi | 78.2 | 3 | 2026-05-30 |
| 3 | Kimi K2.5 | Moonshot | 75.9 | 3 | 2026-05-30 |
| 4 | Kimi K2 (Reasoning) | Moonshot | 75.6 | 3 | 2026-05-30 |
| 5 | DeepSeek-V3.2 | DeepSeek | 75.1 | 3 | 2026-05-30 |
| 6 | MiniMax M2.1 | MiniMax | 64.3 | 3 | 2026-05-30 |
| 7 | GLM-4.7 | Z.ai | 61.9 | 3 | 2026-05-30 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.