All-Angles Bench — Published multi-view/3D spatial-reasoning VLM benchmark (2025); raws fold in spacing variant
| # | Model | Vendor | Best score | Runs | Last seen |
|---|---|---|---|---|---|
| 1 | Step3 VL 10B | StepFun | 57.2 | 2 | 2026-06-05 |
| 2 | HY-Embodied 0.5 MoT-2B | Tencent | 55.1 | 1 | 2026-06-15 |
| 3 | GLM-4.6V-Flash (9B) | Z.ai | 53.2 | 2 | 2026-06-05 |
| 4 | MiMo VL RL 2508 (7B) | Xiaomi | 51.6 | 2 | 2026-06-05 |
| 5 | MiMo Embodied 7B | Xiaomi | 49 | 1 | 2026-06-15 |
| 6 | Qwen3 VL 4B (Reasoning) | Alibaba | 46.7 | 1 | 2026-06-15 |
| 7 | Qwen3 VL Thinking (8B) | Alibaba | 45.9 | 2 | 2026-06-05 |
| 8 | InternVL-3.5 (8B) | OpenGVLab | 45.3 | 2 | 2026-06-05 |
| 9 | Qwen3 VL 2B | Alibaba | 42.3 | 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.