Kakao open-sources Kanana-2-v-decide, a multimodal decision model for agents
On November 6, Kakao said it open-sourced Kanana-2-v-decide on Hugging Face (collection includes kanana-2-v-decide-3b-2611). Built on Kanana-2-v-3b, the multimodal model reads text and images and returns probability, choice, or score outputs instead of long prose, so agents can wire the result into tool choice or support routing. Kakao says average latency is about 0.04 seconds on one H200 GPU and self-reports roughly 43× faster than GPT-6 Luna; it also cites its Kanana Decision Bench and the public Decision Index versus Decision-1, Jev, pplx-decider, and Solar Decide. The weights use the commercial Kanana Open License.
Why it matters
Many agent steps need structured decisions, not paragraphs. A commercially licensed open decision model lets builders skip parsing free-form replies.
Sources2
Written by dalao.ai: the title, summary and why it matters are our own words. The original reports are the record; their publishers keep authorship.