Informe de inteligencia

Sqwish raises $3M pre-seed to let customers tune its AI decision model

The U.K. startup behind the kettei fat decision model is betting that ongoing training on customer data will set it apart in a fast-crowding field.

INSIGHTEX Editorial Desk

2 min de lectura

Sqwish raises $3M pre-seed to let customers tune its AI decision modelInforme de inteligencia

What happened

U.K.-based AI startup Sqwish has raised a $3 million pre-seed round for kettei fat, its decision model, TechCrunch reported. Decision models are LLMs that do not generate output tokens. Instead they choose among preset options and report how confident they are in the pick, which makes them suited to quick, low-cost calls in production systems.

Co-founders Ushnish Sengupta, the CEO, and Federica Freddi met as graduate students at Cambridge. They were building small LLMs when TypeSafe released its Jev models, and concluded that their architectures could be adapted into a decision model. Sqwish's distinguishing feature is post-training tooling built into the product, which lets customers use their own data to sharpen the model's judgement.

Sengupta said this ongoing customization is what keeps Sqwish from becoming a commodity, because customers can keep adding production data and improve their models on the work that matters to them. He described benchmark results, where kettei fat has done well on some tests, as a reason to try the model, and continual tuning as the reason to keep using it.

Context

Decision models appeared only a month before October, when they took more than one in four requests on Vercel's AI model router. TypeSafe became a $7.5 billion company after launching Jev, and OpenRouter handled more than 1.5 billion queries to its models last week. Microsoft CEO Satya Nadella said Microsoft-Decision-1.1 is rolling out for wide use on Azure.

OpenRouter profile

FuenteTechCrunch

El boletín

La semana tecnológica, en un solo correo.

Las crónicas que merecen su tiempo, cada semana. Cancele la suscripción cuando quiera.