Today we release Mirage: A 22 MB visual-language model that runs scene descriptions +…
This is a dev post classified by AGK Intelligence as Other (a launch), kept by the Dev Radar because it carries real work, not commentary.
Today we release Mirage: A 22 MB visual-language model that runs scene descriptions + object detection + OCR + segmentation + spatial relations analysis + extractions + decisions, in a single pass. It can do all that in 74 ms on Samsung Galaxy S26 NPU or 107 ms on its CPU. Mirage has a 7.8M encoder + a 40.8M decoder at 2bits, and describes images in English, German, French, Spanish, Italian, Dutch and Polish. We trade general language capacity, to compete top small VLMs today (SmolVLM2-500M, LFM2.5-450M and Florence2-230m) on these tasks. On 256 VQAv2 questions, feeding Mirage's structu
Posted by Cactus Compute (4.1k followers) 2 h ago · 240 likes · 5.9k views · view the original post on X. Kept by the Dev Radar as Other. Tools mentioned: Cactus Compute.
More dev work like this
- Building for Meta VR Glasses or bringing an existing Quest experience over? 🥽 — @unitygames
- The Malleable Computer, Made in America — @stevederico
- Sei is cutting its code down hard on the road to Giga — @BSCNews
- Across the Brooklyn 6G Summit show floor, partners are putting NVIDIA technology to work… — @nvidianewsroom
- On Canton, scaling into dedicated infrastructure means adding capacity while staying… — @CantonNetwork
- The way these guys are doing streaming speech recognition is really cool. — @svpino
- @Enter_Elysium launches Monday Nov 9th with @Zerion as its preferred data provider. — @zerion
- 1847 Gaussian Machine Learning. #BigData #Analytics #DataScience #MachineLearning #IoT… — @gp_pulipaka
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 31.8k posts from 5k X accounts over the last 21 days, 9.5k tools, 12 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-06 19:42 UTC. Full method.