SSyncDev

Kanana-2-v-decide-3b-2611

SyncDev model directory · Profile updated Nov 18, 2026

Compare with another model

Kanana-2-v-decide-3b-2611 is Kakao's open 3.67B multimodal decision model for Korean and English, scoring choice, yes/no and rating questions.

Key specifications

Specification score
22.8
Context window
20.48K
License
Kanana Open License

The specification score summarizes recorded features and limits. It is not a benchmark result.

Pricing conditions: Open weights; no per-token price is published. Your cost is the GPU time you run it on.

Inputs and outputs

Recorded modality support shows what this model can accept and produce. An undocumented modality is shown as unknown rather than assumed to be unsupported.

Accepts

  • Text input
  • Image input

Produces

  • Embeddings output

Capabilities

VisionOpen weights

What is Kanana-2-v-decide-3b-2611?

Plenty of production LLM calls are not really generation. They are decisions: is this review abusive, which team should get this ticket, how urgent is this message? Kanana-2-v-decide-3b-2611 is Kakao's model for exactly that, and it never writes text. It scores the candidate answers and returns a probability for each.

How it decides

The input, called the state, can be text, a JSON object or array, or a chat history, with up to 10 images. Each question is embedded together with the state into one vector; each candidate answer gets its own. A softmax over cosine similarities produces the distribution.

Three question types:

  • choice — pick one of several options.
  • noul — yes or no, returned as the probability the statement is true.
  • score — a rating on an ordered scale of 2 to 10 levels, returned as an expected level.

Ask any number of questions per request. Remember that probabilities are relative to the options you supply: change an option and the whole distribution shifts. A temperature setting flattens or sharpens it.

Specification

Item Value
Parameters 3.67B (0.49B vision encoder, 3.18B language model)
Built on Kanana-2-v-3b
Embedding size 2,560
Input limit 20,480 tokens, up to 10 images
Released 6 November 2026
Licence Kanana Open License

Kakao reports 72.50 on its in-house Korean decision benchmark, 18 points ahead of the next model it tested, and a 61.23 average across Decision Index and Decision Index Vision, ahead of Liquid AI's d1-3B (see the Liquid AI page) and others in its comparison.

Serving

A bundled script exposes a System One-compatible HTTP API using only the Python standard library. It handles one request at a time and binds to localhost by default; set an API key before exposing it.

Testing it on a real queue

Export five hundred labelled items from a classification flow you run today, ideally Korean and English mixed. Phrase each label as a choice question and compare accuracy and latency with your current LLM prompt. Then check calibration: of the answers it gives 0.9 confidence, roughly nine in ten should be right.

Similar closed-set models are grouped under structured decisions.

How to evaluate Kanana-2-v-decide-3b-2611

Compare the model on the work you need to complete.

Tasks and inputs

Kanana-2-v-decide-3b-2611 is listed for Zero-shot classification, Structured Decisions. It accepts text, image and produces embedding. Use the publisher’s input format and preprocessing for your chosen deployment.

Capacity and cost

Documented limits: 20.48K context. Evaluate retrieval over a labelled query set. Keep the encoder, vector dimensions and index settings consistent.

Choosing a deployment

The weights are downloadable under Kanana Open License. Check runtime support and size the complete model and working context for your expected concurrency.

Frequently asked questions

No. It returns a probability for each candidate answer to choice, yes/no or score questions.

These records share a directory category. Check the source card, output type and deployment requirements before treating a model as a replacement.

Microsoft

Microsoft-Decision-1.1

Microsoft-Decision-1.1 is a decision model for routing, triage and classification, returning calibrated probabilities. GA in Foundry from November 2026.

— context

Liquid AI

d1.2

Liquid AI d1.2 is a hosted omnimodal decision model for text, image and audio questions with calibrated answers, at $0.02 per million input tokens.

65.54K context

Liquid AI

d1.2-ultra

Liquid AI d1.2-ultra is the top hosted decision model in Liquid's line, scoring 75 on the AA Decision Index, with audio support at $0.042 per 1M input.

65.54K context

Google

Gemini Decision 1

Gemini Decision 1 powers Google's Decisions API, returning calibrated yes/no, choice and rubric answers instead of text, at $0.10 per 1M input tokens.

262.14K context

subquadratic

SubQ Snap 1.2

SubQ Snap 1.2 is Subquadratic's long-context decision model on linear sparse attention, tested to 256K tokens, priced at $0.04 per 1M input tokens.

— context

Perplexity

pplx-decider-flash-v1.3

pplx-decider-flash-v1.3 is Perplexity's small, half-price decision model, distilled from the 27B decider, for high-volume classification and span extraction.

— context