Pending benchmarksTracking since Oct 19, 2026
Is MJev-Spatial state of the art?Pending benchmarks.Announced Oct 19, 2026; no independent results yet.
Mind Lab's System 1 spatial decision model on Qwen3.5-4B: scores answer options, value bins and image regions with native attention Q/K, without autoregressive answer tokens.
Announced facts
The details so far
- Lab
- Mind Lab (Macaron AI)
- Announced
- Oct 19, 2026
- API
- Not yet
- Open weights
- Published
- Parameters
- 4B (Qwen3.5-4B backbone)
- Type
- Multimodal spatial decision model (choice, numerical, pointing)
- Base
- Qwen3.5-4B; reuses pretrained Q/K projections, no separate prediction head
- Training
- 12.3M QA pairs (lab announcement)
- Released
- Paper, code and weights
- License
- Apache-2.0 (badge on the GitHub README)
- Paper
- arXiv 2610.22276
Tags marked unsure are not stated on the lab's own pages.
Next steps
What would let us call it
Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.
- Publicly announced with a source
- Usable by the public (API or weights)
- Listed on an independent board we track
- Enough results for a rank in a category
- Holds across two snapshots (crowns only)
Self-reported
What the lab claims
Self-reported. Shown for context; not used in any verdict.
Lab claim: #1 on 7 of 12 benchmarks at 65.64% overall (best of 8 models), 5.6× faster than the base model (self-reported).
Receipts
Sources
LABMacaron / Mind Lab announcement on Xx.com/Macaron0fficial/status/2112182654552400249 · Oct 19, 2026PAPERMJev-Spatial paper (arXiv 2610.22276)arxiv.org/abs/2610.22276 · Oct 19, 2026REPOMJev-Spatial-4B on Hugging Facehuggingface.co/mindlab-research/MJev-Spatial-4B · Oct 19, 2026REPOFr0zenCrane/MJev-Spatial on GitHubgithub.com/Fr0zenCrane/MJev-Spatial · Oct 19, 2026Tracking continues until an independent board lists it.