Barely two weeks after TypeSafe AI launched Jev, OpenAI, Amazon Web Services, and Cloudflare each released their own decision models. These systems don’t generate text. They select an answer from a fixed set of options and attach a probability to it. The flurry of releases has turned what started as a niche concept into a crowded field.

TypeSafe calls these “System One” models. They handle the quick, everyday judgment calls AI agents face constantly—deciding if a task is urgent, whether to keep going or stop, or whether to allow or block an action.

TypeSafe Sets the Template with Jev

TypeSafe AI was founded by Diogo Almeida, a former OpenAI researcher. The company released Jev in mid-September and frames its output as calibrated decisions rather than text. In the launch post, Almeida explained that the models produce decisions with probabilities and confidence instead of words.

TypeSafe says Jev matches existing large language models on these tasks while running two orders of magnitude faster. Pricing sits at $0.042 per million tokens. The company emerged from stealth with $40 million in seed funding led by DCVC and has drawn investor interest at a $10 billion valuation. TypeSafe has not disclosed Jev’s architecture.

OpenAI’s Limited Decisions API

At its developer event in late September, OpenAI introduced a limited release of a Decisions API built on its Luna model. CEO Sam Altman noted that narrowing the range of choices delivers very fast speed while still keeping image understanding, broad language support, and safety guardrails.

Amazon’s Open-Source Strands Decider 2B

Amazon’s Strands Agents team released Strands Decider 2B as an open-source model under the Apache 2.0 license. It runs on a local CPU or GPU and returns answers in tens of milliseconds. Built on Qwen3.5-2B, it delivers choices, yes/no probabilities, or scores in a single forward pass.

AWS noted that developers already rely on small decision models for work like model routing, tool selection, guardrails, and policy classification.

Cloudflare’s Clef and Clef-Flash

Cloudflare released two open-weight models, Clef and Clef-flash, hosted on its Workers AI platform. The company also launched a new reinforcement-learning fine-tuning service. Cloudflare states that Clef outperforms Jev on accuracy and latency across dozens of benchmarks. Clef-flash shows a median latency of about 39 milliseconds.

On Workers AI, Clef costs $0.24 per million tokens—roughly six times Jev’s price. Users can still download the weights for free.

Commoditization Pressure on TypeSafe

The open releases create pressure on TypeSafe’s business model. One analysis described the shift as commoditization by subsidy: a startup sells API access while two of the largest infrastructure companies release equivalent capability under a free license.

Researchers have released dozens of similar models since TypeSafe introduced the approach. VentureBeat labeled the AWS release a “Jev killer.” TypeSafe CEO Diogo Almeida has said he does not yet see real competition, telling TechCrunch that people may be underestimating how hard it is to make the models actually smart.