By No Code MBA · Facts checked
- Provider
- Sqwish Labs
- Announcement date
- Status at last check
- Generally available
- Model family
- kettei
- Capabilities
- Text, Image
What changed
kettei is a family of decision models: given a context, a question and a list of allowed answers, they return a probability for each answer in one pass, for uses like tool selection, answer checking, routing and triage. kettei fat and chub accept text and images; core (3.7B) and dot (900M) are text-only and built on IFM's K2-Horizon models. Sqwish reports fat at 97.5% on a 5,000-question held-out image set and about 26 ms of GPU time for fat on a short request.
Access and availability
kettei fat, chub and core are available through the OneHot API, with fine-tuning live for chub and core and in early preview for fat. core and dot are open source, but the post gives no repository link.
Official access documentationWhat you can build
Try an exact, bounded decision task such as routing or evidence checks with fixed choices, then compare its scores and review workload against the current classifier or LLM workflow.
Limitations
The models choose among supplied options and do not write open-ended text. Sqwish says frontier LLMs remain more accurate on its Decision Index and that intent routing is a weak spot, and most benchmarks are vendor-run, some on its own test sets.