# Decision models move from niche to infrastructure as Microsoft rolls out Decision-1.1 and Sqwish raises $3 million

Published 2026-11-17T18:24:57+00:00 · Updated 2026-11-17T18:24:58+00:00 · By superintelligencenews@gmail.com
Canonical: https://superintelligencenews.com/ai-fields/large-language-models/decision-models-microsoft-rollout-sqwish-funding/

> Decision models are moving into the enterprise mainstream, with Microsoft rolling out Decision-1.1 on Azure and Sqwish raising $3 million to focus on customer-specific tuning. Usage data from Vercel and OpenRouter suggests the category is already seeing significant demand.

## Key facts

- In October, more than 25% of Vercel AI model router requests went to decision models, according to TechCrunch.
- Microsoft began a wide Azure rollout of Microsoft-Decision-1.1 on Tuesday.
- Sqwish announced a $3 million pre-seed round today and says continual tuning on customer production data is its key differentiator.
- OpenRouter handled more than 1.5 billion queries to TypeSafe models last week.
- TypeSafe’s Jev family drove the company to a $7.5 billion valuation.

Decision models are moving out of the novelty bucket and into mainstream enterprise infrastructure, with Microsoft beginning a broad Azure rollout of Microsoft-Decision-1.1 on Tuesday and startup Sqwish announcing a $3 million pre-seed round today. The shift matters because demand for these fast, cheap, choice-making systems is rising quickly, and companies now appear to be competing less on raw benchmark scores than on how well they can calibrate judgment for real-world use.

 

## Decision models are gaining traction fast

 

TechCrunch reported that in October, more than 25% of Vercel AI model router requests went to decision models, a sign that these systems are no longer a side category in the AI stack. The outlet describes decision models as LLMs built for fast, inexpensive option selection, often with confidence scoring attached.

 

That framing helps explain why the category is drawing attention from both hyperscalers and startups. A model that can sort choices, classify inputs or make a constrained decision quickly is not trying to replace every general-purpose chatbot. Instead, it is designed to take a narrow but commercially important slice of enterprise work and do it cheaply and reliably.

 

### What makes decision models different?

 

Decision models are essentially tuned for judgment under constraint. Rather than generating open-ended prose, they are used to pick among options, score outcomes or guide a workflow where speed and cost matter as much as raw intelligence.

 

That distinction is becoming more important as businesses compare what models can do in a benchmark suite with what they can do in production, where latency, calibration and failure rates matter more than abstract scores.

 

## Microsoft pushes Decision-1.1 into Azure

 

Microsoft said it began a wide rollout of Microsoft-Decision-1.1 on Azure on Tuesday. According to TechCrunch, the company is positioning the model as a system that balances cost, latency, intelligence and calibration — four attributes that enterprise buyers tend to care about when the model is embedded in a live workflow.

 

Microsoft engineering VP Achint Srivastava said that “benchmaxed models fall flat under real enterprise usage very quickly,” a blunt reminder that enterprise AI procurement is often less about leaderboard prestige than about stability in production.

 

Microsoft also said Decision-1.1 and TypeSafe’s Jev perform strongly across four dimensions on held-out benchmarks. That matters because the company is trying to show that practical value and benchmark strength do not have to be in conflict, even if benchmarks are only one piece of the decision.

 

### Why Microsoft is leaning in

 

The Azure rollout suggests Microsoft sees decision models as a meaningful layer in its enterprise AI stack, not merely a niche add-on. In cloud terms, a model that can be integrated into customer systems and tuned to specific workflows creates recurring value, particularly if it lowers costs while improving speed and confidence in outputs.

 

It also puts Microsoft in a race that is not only about model quality but about packaging. If decision models become standard enterprise building blocks, the winner may be the company that makes them easiest to deploy, monitor and adapt.

 

## TypeSafe’s Jev family shows the scale of demand

 

TypeSafe has become one of the clearest signs that this market is growing. TechCrunch reported that the company’s Jev family drove it to a $7.5 billion valuation, and OpenRouter handled more than 1.5 billion queries to TypeSafe models last week.

 

Those figures suggest a meaningful appetite for specialized decision systems rather than a brief experimental spike. They also point to a market where usage volume can climb fast once a model proves useful in production settings.

 

In TechCrunch’s account, TypeSafe CEO said the hardest part was calibrating judgment with statistically adjusted synthetic data. That detail goes to the heart of the category: decision models may be simpler than frontier chat models in some respects, but getting them to behave correctly in edge cases is still technically demanding.

 

### By the numbers

 

| Metric | Figure | What it refers to |
| --- | --- | --- |
| Vercel AI router share | More than 25% | October requests going to decision models |
| TypeSafe valuation | $7.5 billion | Company value after Jev models gained traction |
| OpenRouter query volume | More than 1.5 billion | Queries to TypeSafe models last week |
| Sqwish pre-seed round | $3 million | Funding announced today |
| Freddi’s opportunity estimate | $10 billion to $50 billion | System One intelligence market range |
| OpenAI comparison | $1 trillion to $2 trillion | Market value range cited by Freddi |

 

## Sqwish bets that customization is the real moat

 

Sqwish announced a $3 million pre-seed round today as it tries to carve out room in the same emerging category. The company’s model is called kettei fat, and its founders, Ushnish Sengupta and Federica Freddi, are Cambridge graduate students who first worked on small LLMs before pivoting to decision models after the launch of Jev.

 

Sqwish says it is using synthetic data together with customer-data post-training tools to improve judgment, and it argues that continual tuning on customer production data is its key differentiator. That is a notable bet in a market where large companies can easily copy broad product ideas, but may not replicate the same level of workflow-specific customization.

 

Sqwish is positioning itself not as a generic model provider but as a specialist in making decision systems more useful inside customer environments. The company’s pitch is that the model is only part of the product; the real advantage is the ongoing adjustment layer around it.

 

### What Sqwish thinks the market looks like

 

Freddi, according to TechCrunch, sees decision models as the “system one” layer of AI, paired with “system two” frontier LLMs. That is a useful way to frame the opportunity: one class of model handles fast, routine judgment, while a second class handles deeper reasoning or more open-ended tasks.

 

Freddi also estimated that the System One intelligence market could be worth between $10 billion and $50 billion, far below the market value OpenAI is said to be targeting, which she compared with a range of $1 trillion to $2 trillion. The comparison underscores how startup founders are trying to define a large but distinct market beneath the blockbuster frontier model race.

 

## Why it matters

 

This story matters because decision models may become a practical layer of enterprise AI infrastructure, not just a technical curiosity. If the category keeps expanding, businesses will increasingly buy systems that are optimized for speed, cost and calibrated judgment rather than for broad conversational ability.

 

That shift has several implications. It could change how cloud providers package AI services. It could also redirect startup funding toward model customization, post-training tools and workflow integration. Most important, it may show that the next phase of AI competition is not only about building the biggest model, but about making a model trustworthy enough to use at scale.

 

For readers, the commercial significance is straightforward: if decision models perform well, they can reduce the cost of automating decisions that currently require human review. For the market, the stakes are bigger. A useful decision layer could become embedded in support systems, approval flows, routing engines and other enterprise tools that process huge volumes of routine choices.

 

## Background: how the category got here

 

Decision models fit into a broader shift in AI toward systems that are specialized rather than universal. The fact sheet describes them as LLM-derived tools optimized for fast classification or choice among preset options. That makes them different from general-purpose assistants, even if they are built from related technologies.

 

The timing also matters. Synthetic data has become a standard tool in AI training when real labeled data is limited, and the current market has been shaped by a broader recognition that benchmarks do not always predict enterprise performance. In practice, companies care about reliability, speed, cost and calibration in workflows that must work repeatedly, not just once in a demo.

 

That tension between lab performance and production reality is at the center of the current competition. Microsoft is stressing calibration and enterprise fit. TypeSafe’s rapid usage suggests there is already demand for systems built around narrow decisions. Sqwish, meanwhile, is making its case on continuous tuning and customer-specific data.

 

There is also a competitive dynamic between large cloud providers and startups. Microsoft can distribute a model through Azure at scale. Startups can move faster on customization and tight product design. The emerging market may reward both, but in different ways.

 

## What the usage data suggests

 

The usage numbers in the fact sheet point to a category that is moving quickly from testing to deployment. More than 25% of Vercel AI model router requests going to decision models in October suggests meaningful adoption among builders. OpenRouter handling more than 1.5 billion queries to TypeSafe models last week suggests serious volume, not isolated experimentation.

 

Those figures do not prove the category’s long-term size, but they do show momentum. They also indicate that infrastructure providers are likely to pay close attention to routing patterns, latency profiles and cost curves as more enterprise traffic flows through these systems.

 

### Usage and rollout timeline

 

| Date | Event | Significance |
| --- | --- | --- |
| October | More than 25% of Vercel AI router requests went to decision models | Showed demand was already significant |
| Last week | OpenRouter processed more than 1.5 billion TypeSafe queries | Highlighted production-scale usage |
| Tuesday | Microsoft began a wide Azure rollout of Microsoft-Decision-1.1 | Put a major cloud provider into the category |
| Today | Sqwish announced a $3 million pre-seed round | Added a fresh startup bet to the market |

 

## How are startups trying to compete?

 

Startups are trying to compete by narrowing the problem and deepening the tuning. Sqwish’s answer is that continual tuning on customer production data can create a stronger product than a one-time benchmark win. That is a direct challenge to the idea that model quality can be measured once and then treated as fixed.

 

TechCrunch reported that Sengupta described the company’s edge this way: “What makes us not a commodity, I think, is the fact that we are offering our customers the same power…” The line reflects a broader startup strategy: avoid being viewed as a generic model vendor and instead sell a combination of base capability and customer-specific adaptation.

 

Sqwish’s cofounder also sees a world in which decision models pair with frontier LLMs rather than replace them. That division of labor could help explain why the market is opening up at all: a broad AI stack may need different kinds of systems for different classes of tasks.

 

## What happens next

 

The next phase will be less about launch announcements and more about deployment evidence. The key open questions are whether decision-model demand keeps rising beyond October’s usage spike, whether Microsoft-Decision-1.1 performs well in customer environments, and whether Sqwish can maintain a defensible customization advantage against much larger rivals.

 

Watch for three things in particular:

 

- Whether enterprise customers report strong real-world performance from Microsoft-Decision-1.1.
- Whether routing traffic continues to tilt toward decision models on platforms like Vercel and OpenRouter.
- Whether Sqwish’s post-training tools and customer-data tuning become enough of a product moat to attract more customers and investors.

 

Also worth watching is which use cases prove most durable. The fact sheet does not name them, but the strongest adoption is likely to come from repeated, high-volume decisions where speed, consistency and cost control matter more than open-ended generation. If that pattern holds, decision models could become one of the more important operational layers in enterprise AI.

 

## Bottom line

 

Decision models are emerging as a real category, not a side experiment. Microsoft is pushing one into Azure, TypeSafe is already seeing scale, and Sqwish is betting that customer-specific tuning will matter more than generic model bragging rights. The competition now looks less like a race to build the smartest all-purpose model and more like a contest to make fast judgment reliable enough for everyday business use.

## Frequently asked questions

### What are decision models in AI?

Decision models are AI systems built for fast, cheap choice selection rather than open-ended text generation. TechCrunch describes them as LLMs optimized for classification, option picking and confidence scoring in workflows where speed, cost and calibration matter.

### Why is Microsoft rolling out Decision-1.1?

Microsoft is rolling out Decision-1.1 on Azure because enterprise customers care about cost, latency, intelligence and calibration. The company is positioning the model as a practical tool for real workflows, where benchmark performance alone does not guarantee usefulness.

### How much money did Sqwish raise?

Sqwish raised $3 million in a pre-seed round announced today. The startup says it will use synthetic data and customer-data post-training tools to improve the model’s judgment and keep tuning it on production data.

### Why are startups interested in decision models?

Startups are interested in decision models because they may offer a clearer path to enterprise value than broad, general-purpose AI. The category rewards customization, calibration and workflow integration, which can be easier to sell into businesses than raw benchmark scores.

## Sources

- [TechCrunch](https://techcrunch.com/2026/11/17/as-decision-model-use-skyrockets-is-customization-is-the-next-frontier/)
