Stop using frontier LLMs for vision: a $6 detector trainer, distilled into agent tools

mervenoyann · x · 2026-11-17

mervenoyann argues you shouldn't use frontier LLMs as vision interns for detection/OCR. Instead she packaged her favorite vision models (keypoint detection to OCR) as agent-callable tools, plus an end-to-end object detection trainer covering labelling, VLM-as-judge, training and eval — a detector for $6 that runs cheap. Talk from AI Engineer WF is out; GitHub repo coming with decision model support and updated VLMs (LFM3, Qwen3.6/8).

Related event: Hugging Face Open-Sources vision-intern: Visual Tools for Agents at $6(2 posts)→

Original post →

More from coding & agent

coding & agent channel →