cd /news/artificial-intelligence/keel-a-frontier-model-teaches-a-visi… · home › topics › artificial-intelligence › article
[ARTICLE · art-173032] src=arxiv.org ↗ pub= topic=artificial-intelligence verified=true sentiment=↑ positive

Keel: A Frontier Model Teaches a Vision-Language Model by Rewriting Its Weights

A new arXiv paper (2611.25257v1) introduces Keel, described as the first agentic weight-editing method for vision-language models on image tasks, in which a frontier multimodal LLM acts as an editor that locates the cause of looping behavior in a VLM's weights, edits them and tests the result without any training. On Qwen3.5-0.8B and Qwen3.5-2B across six image benchmarks from visual mathematics to GUI grounding, the edit removes most loops on every benchmark, shortens answers, and the edited models are significantly more accurate on average than the thinking models, with the larger model gaining most on GUI grounding, a task absent from the practice questions. The editors traced the loops to the attention outputs of the middle layers, where the smaller VLM stops consulting the image inside a loop, and Keel combines with budget forcing to give the best average accuracy on both models, significantly above every training-free baseline compared on the larger one.

recommendation

Gate agent-proposed weight edits on held-out questions: keep only edits that clearly reduce the failure without lowering accuracy on questions no editor saw, as Keel's harness does.

more in playbook: Evals and testing for LLM apps
by read2 min views2 publishedNov 20, 2026

arXiv:2611.25257v1 Announce Type: new Abstract: Can a frontier model repair a vision-language model (VLM) by rewriting its weights? Small reasoning VLMs such as Qwen3.5 need such a repair: in thinking mode they fall into loops on many image questions, repeating themselves until the token limit, spending the whole budget and losing the answer. The usual remedies act on the decoding, leaving the looping model as it was, or must be told what to change. Yet the cause is written in the weights. Keel, the first agentic weight-editing method for VLMs on image tasks, finds and removes it there. Told only the symptom, an editor, a frontier multimodal LLM acting as an agent, locates the cause, edits the weights and tests them, with a failure census and instruments showing it the failing images and where the VLM attends; among several editors, the harness keeps only edits that clearly reduce the failure without lowering accuracy on questions no editor saw. The repair needs no training, keeps the model's size and per-token cost, and applies to already-edited weights. The editors expose where the loops live: from the original weights, independent editors on both models trace them to the attention outputs of the middle layers, and the smaller VLM stops consulting the image inside a loop. On Qwen3.5-0.8B and Qwen3.5-2B, across six image benchmarks from visual mathematics to GUI grounding, the edit removes most loops on every benchmark and shortens answers, and the edited models are significantly more accurate than the thinking models on average; the larger one gains most on GUI grounding, a task absent from the practice questions. Because the repair lives in the weights, it combines with budget forcing where held-out questions favor it: on the best available weights, Keel has the best average accuracy on both models, significantly above every training-free baseline we compare on the larger one.

── more in #artificial-intelligence 4 stories · sorted by recency
── more on @keel 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/keel-a-frontier-mode…] indexed:0 read:2min 2026-11-20 · —