Decision models grab a quarter of Vercel's AI traffic a month after Jev
Microsoft is pushing its own decision model into wide release on Azure, while UK startup Sqwish just raised $3 million to make them customizable. The category barely existed a month ago, yet decision models already handle more than one in four requests on Vercel's AI router.
- Original source: techcrunch.com.
- The source published it on Nov 17, 2026 at 6:10 PM.
- No second outlet has reported it yet. Treat it as a single-source claim.
Microsoft CEO Satya Nadella said Tuesday that the company is rolling out Microsoft-Decision-1.1 for broad use on Azure, effective immediately. The model is the product of balancing cost, latency, intelligence and calibration, according to Achint Srivastava, the engineering VP who led the project. Srivastava warned in an email that models tuned purely for benchmarks fall apart fast in real enterprise work, and said Microsoft-Decision-1.1 and TypeSafe's Jev are among the few that hold up across all four dimensions, including on unseen tests built to mirror real use.
Also Tuesday, UK startup Sqwish announced a $3 million pre-seed round for its decision model, kettei fat. Cofounders Ushnish Sengupta and Federica Freddi met as Cambridge grad students building small LLMs, then spotted that their architecture could be reworked into a decision model once Jev landed. Sqwish leans on synthetic data like its rivals, but its pitch is built-in post-training tools that let customers keep feeding in their own production data so the model sharpens on the tasks they care about.
The demand is real. More than one in four requests on Vercel's AI model router went to decision models in October, barely a month after the concept existed. Instead of generating output tokens, these models spend their intelligence cheaply and quickly picking between preset options and reporting how confident they are — a fit for production systems where speed beats the slow reasoning of a pricey frontier model. TypeSafe became a $7.5 billion company on the back of Jev, and OpenRouter fielded more than 1.5 billion queries to TypeSafe's models last week.
TypeSafe CEO Diogo Almeida told TechCrunch the hardest part of building Jev was fine-tuning its judgement with statistically-adjusted synthetic data. That barrier is low for everyone else too: dozens of decision models shipped after Jev, including one an Amazon engineer assembled over a weekend, and every hyperscaler now has an option. Ranking them is messy — most builders say the only real test is swapping one model into a specific workflow and watching what happens.
Sengupta frames the durability of his company as the selling point. Benchmarks, he said, are an invitation to try kettei fat, while continual tuning is why customers stay. Freddi expects decision models to eventually run alongside frontier LLMs, pairing fast System One thinking with slower System Two reasoning — and she sized the standalone opportunity at $10 billion to $50 billion.
Even in that split world, Freddi argues the giants won't fight hard for it, pointing to OpenAI's ambitions of a $1 trillion to $2 trillion market value around an IPO. Decision models, she told TechCrunch, are all Sqwish cares about, while for the frontier labs it's a side project.
TopicsMicrosoft · Azure · Satya Nadella · Sqwish · Ushnish Sengupta · Federica Freddi · TypeSafe · Vercel
Source: techcrunch.com
Written by AI from verified reporting.
More stories
- Tech & AI cryptobriefing.com 0 viewsAnthropic IPO order book shuts at 4 p.m. ET, reportedly six times oversubscribed
- Tech & AI breakingthenews.net 0 viewsNvidia Taps Heidrick & Struggles to Run Its CEO Search
- Tech & AI x.com 0 viewsOpenAI launches GPT-6.2 Astra and Sol, billed as its most powerful models yet
- Tech & AI x.com 0 viewsAnthropic eyes IPO price above $80 as demand blows past range

