Editorial illustration for UK AI startup Sqwish raises USD 3M pre-seed for decision model kettei fat
Sqwish Raises $3M for Decision Models Over LLMs
UK AI startup Sqwish raises USD 3M pre-seed for decision model kettei fat
Vercel's AI model router logged something strange in October: more than one in four requests went to decision models instead of large language models. That category didn't exist a month before. Decision models skip output tokens entirely.
They're tuned to pick between preset options fast and cheap, attaching a confidence score to each choice rather than reasoning through an answer. For production systems where speed beats explanation, that trade-off sells itself.
TypeSafe set the pace, hitting a $7.5 billion valuation after shipping its Jev model family, which OpenRouter says still handled over 1.5 billion queries last week alone. But the technique turned out to be easy to copy. One Amazon engineer built a working decision model in a weekend. Every major hyperscaler followed with its own version, and on Tuesday Microsoft CEO Satya Nadella announced Microsoft-Decision-1.1 is rolling out on Azure immediately.
Now the race is shifting from who can build a decision model to who can build a better one. UK startup Sqwish, founded by Cambridge grad students Ushnish Sengupta and Federica Freddi, just raised $3 million pre-seed on exactly that bet.
“What makes us not a commodity, I think, is the fact that we are offering our customers the same power…they can keep uploading their production data, and just the same way that we trained our models, their models can also keep getting better at the tasks that they care about,” CEO Ushnish Sengupta told TechCrunch.
Why this matters
A category invented a month ago already accounts for more than a quarter of traffic on Vercel's router. That pace should give us pause, not applause. Decision models are cheap and fast precisely because they're narrow, and narrowness is exactly what invites a customization land grab.
Sqwish's $3 million pre-seed for kettei fat tells us investors are betting the next competitive edge isn't bigger models but better-tuned small ones built for specific production decisions. Sengupta and Freddi's background building small LLMs before pivoting suggests this isn't a bolted-on pivot chasing a trend, it's a team that saw the infrastructure need coming.
For developers and founders, the practical question is whether decision models become a commodity layer everyone plugs in, or whether companies like Sqwish carve out defensible ground through customization tooling. For researchers, the confidence-scoring mechanism deserves scrutiny: how these models calibrate certainty will matter more than raw speed once they're embedded in real production pipelines. Watch whether other startups follow Sqwish into the customization layer, or whether the big model providers absorb this functionality themselves.
Common Questions Answered
What are decision models and how do they differ from large language models?
Decision models skip output tokens entirely and are tuned to pick between preset options quickly and cheaply, attaching a confidence score to each choice rather than reasoning through a full answer. Unlike large language models that generate explanations, decision models prioritize speed and cost efficiency for production systems where fast decision-making is more important than detailed reasoning.
Why did Vercel's AI model router see such a significant shift toward decision models in October?
More than one in four requests on Vercel's router went to decision models in October, despite this category not existing a month prior. This rapid adoption occurred because decision models offer a compelling trade-off for production systems: they deliver faster results at lower costs compared to traditional large language models.
What is Sqwish's competitive advantage with kettei fat according to CEO Ushnish Sengupta?
Sengupta stated that Sqwish differentiates itself by allowing customers to continuously upload their production data, enabling their models to keep improving at the specific tasks they care about. This customization capability prevents Sqwish from becoming a commodity offering in the decision model space.
Why are investors betting on customized decision models as the next competitive frontier?
Decision models are inherently narrow and cost-effective, which creates an opportunity for customization-focused companies to capture market share. Sqwish's $3 million pre-seed funding reflects investor confidence that the next competitive edge lies in better-tuned small models built for specific production decisions rather than larger, more generalized models.
Further Reading
- As decision model use skyrockets, is customization the next frontier? - TechCrunch
- Decision models take 26% of requests; Anthropic now wins new teams 2:1 - Vercel
- Perplexity's Decider V1.2 models now available on AI Gateway - Vercel
- Laya decision model now available on AI Gateway, free through October 31 - Vercel