Model surgery
Giving open models eyes
Freeze the text backbone, freeze an existing vision tower (Kimi K2.6’s MoonViT), and train only a small projector between them. WebBrain did this for Poolside’s Laguna XS 2.1 and for DeepSeek V4 Flash, and published the results.
Why it matters
The cheapest strong planners are often text-only, and text can’t show a canvas, a chart or an overlay covering a button. A graft adds sight for a tiny fraction of the cost of training a multimodal model.
Laguna XS 2.1 Vision
Poolside’s 33B-total, 3B-active mixture-of-experts gets a frozen 416.9M-parameter MoonViT and a newly trained 30.68M-parameter projector, trained on 100,000 examples with up to 512 image tokens. In text-only planning it edges Qwen 3.6 27B on exact-action consensus (41.3% vs 38.1%), twice as fast (1.15 s vs 2.23 s median) at about a ninth of the cost per replay ($0.073 vs $0.670). Qwen still leads on exact ideal actions, 17 to 5.
DeepSeek V4 Flash Vision
The same approach adds vision to the previously text-only DeepSeek V4 Flash, released in NVFP4 and BF16, including the 0731 revision.
Models
-
Laguna-XS-2.1-Vision-NVFP4Poolside Laguna XS 2.1 with MoonViT vision, NVFP4 104 downloads · 3 likes -
Laguna-XS-2.1-Vision-BF16BF16 18 downloads · 0 likes -
DeepSeek-V4-Flash-Vision-NVFP4DeepSeek V4 Flash with vision, NVFP4 208 downloads · 20 likes -
DeepSeek-V4-Flash-Vision-BF16BF16 0 downloads · 0 likes -
DeepSeek-V4-Flash-0731-Vision-NVFP4The 0731 revision with vision, NVFP4 106 downloads · 33 likes -
DeepSeek-V4-Flash-0731-Vision-BF16The 0731 revision with vision, BF16 160 downloads · 6 likes
Caveats
- Laguna XS 2.1 Vision is experimental, not yet a drop-in serving release; serving integration, NVFP4 equivalence and final image-inference validation are pending. Its benchmarked rows are text-only.
- Adding vision to DeepSeek V4 Flash does not change its text-only results.