The kettei family Ready to use or tuned on your own data.

ModelIntelligenceSpeedPriceInputFine-tunableTry it

kettei-fat

Extra high
Very high
$0.06per 1M tokens
Text, image
Early preview

kettei-chub

High
Very high
$0.04per 1M tokens
Text, image
Yes

kettei-core

Medium
Ultra
$0.02per 1M tokens
Text
Yes

kettei-dot

Basic
Ultra
Freeopen weights
Text
On your device

Ahead of every model in their class.

Read the release post →

4050607010 ms30 ms100 ms300 ms1 s3 s↖ FASTER AND MORE ACCURATEMEDIAN LATENCYLiquid d1-3BvLLM SR d3Perplexity Decider v1.1TypeSafe Jev 1.13Microsoft Decision-1OpenAI DecisionsFrontier LLMsdotcorechubfat4050607010 ms30 ms100 ms300 ms1 s3 s↖ FASTER AND MORE ACCURATEMEDIAN LATENCYLiquid d1-3BvLLM SR d3Perplexity Decider v1.1TypeSafe Jev 1.13Microsoft Decision-1OpenAI DecisionsFrontier LLMsdotcorechubfat
Text decisions on the Decision Index 0.3.1. Up and left is better. Frontier LLMs score higher, and take seconds.

Fine‑tune to make it yours.

Teach it your task in minutes, and it beats frontier LLMs at a fraction of their cost.

Legal contract review

Chub beats Claude Opus 5.5 on merger contracts.

+8 pts
accuracy, from 86.5% to 94.5%
200×
cheaper to run than Claude Opus 5.5
20 min
to fine-tune
70%75%80%85%90%95%kettei chub86.5%94.5%+8 POINTSClaude Opus 5.593.6%kettei fat91.8%GPT-6 Sol91.7%TypeSafe Jev 1.1373.8%70%75%80%85%90%95%kettei chub94.5%+8 POINTS86.5%Claude Opus 5.593.6%kettei fat91.8%GPT-6 Sol91.7%TypeSafe Jev 1.1373.8%

Customer support

Chub beats Claude Opus 5.5 on support tickets.

+10 pts
accuracy, from 83.1% to 93.4%
10×
faster than Claude Opus 5.5
15 min
to fine-tune
70%75%80%85%90%95%kettei chub83.1%93.4%+10 POINTSkettei fat89.6%Claude Opus 5.587.3%TypeSafe Jev 1.1381.0%GPT-6 Luna79.4%70%75%80%85%90%95%kettei chub93.4%+10 POINTS83.1%kettei fat89.6%Claude Opus 5.587.3%TypeSafe Jev 1.1381.0%GPT-6 Luna79.4%

Continual learning

A model that keeps learning after it ships.

V1V2V3V4V5 KEEPS LEARNING · BETTER EVERY ROUNDTRAINED ONCE · FALLS BEHIND