MJev-Spatial - RoboSpatial Pointing: leaderboard
MJev-Spatial v1 Table2 final12.3M-QA-trained model from post-trainedQwen3.5-4B; native final-layer Q/K candidate classification, distinct from autoregressive baseline. Eight models on complete twelve-benchmark main comparison; Overall equal benchmark mean. Historical72K appendix runs are different checkpoints and not merged. Source-specific protocols retain Point-Bench five-category macro incl exact-count requirement; Where2Place grayscale-mask scoring; MMSI not circular accuracy. VSI final profile32frames64image tokens/frame. No claim that synthetic repo examples are the exact scored task manifest. Metric: Any-first-two point mask success (%). Higher is better. Source: https://arxiv.org/html/2610.22276v1
Metric: Any-first-two point mask success (%). Source: arxiv.org. Saturation forecast: Around January 2027. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen 3 VL 8B Instruct | 59.02 | #414 |
| 2 | Qwen 3.5 4B | 50 | #469 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=mjev-spatial-robospatial-pointing · How It Works · Data refreshed daily, snapshot 2026-10-17.