SAT (spatial-aptitude embodied benchmark) — Alongside RoboBench/RefSpatial in Tencent HY-Embodied card — spatial bench, not college exam.
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | MiMo Embodied 7B | Xiaomi | 78.7 | 1 | 2026-06-15 |
| 2 | HY-Embodied 0.5 MoT-2B | Tencent | 76.7 | 1 | 2026-06-15 |
| 3 | Qwen3 VL 4B (Reasoning) | Alibaba | 69.3 | 2 | 2026-07-26 |
| 4 | Mage-VL-4B | Microsoft | 67.3 | 1 | 2026-07-26 |
| 5 | Phi 4 R V 15B | Microsoft | 66.7 | 1 | 2026-07-26 |
| 6 | Phi 4 MM 5.6B | Microsoft | 55.3 | 1 | 2026-07-26 |
| 7 | Qwen3 VL 2B | Alibaba | 45.3 | 1 | 2026-06-15 |
Best tracked score per model (default configuration; source-attributed and verification-tiered). Open a model for its full benchmark surface, provenance and pricing.