Jev decision model and paper
What happened
Simon Willison reported on September 21, 2026 that TypeSafe AI had unveiled Jev, a new model category returning probabilistic decisions for classification.1 Willison gave the price as $0.042 per million input tokens, with free output.1 Latent Space reported the same day that the launch reached about 40 million video views and 1 trillion daily tokens in its first days.2
Early tools and adoption
Willison released a plugin, llm-typesafe, that adds Jev support to the LLM CLI tool and Python library on September 22, 2026.3 The plugin supports yes/no, choice and scoring questions.4 Latent Space reported on September 25 that Jev costs $0.044 per 1K judgments, 277x cheaper than GPT-6 Astra.5 Nate B Jones reported on September 27 that Jev reached 2x more paid teams than previous launches on Vercel's AI Gateway in its first 24 hours.6
A paper and a rival model
Latent Space reported on September 29 that a paper called "Jev as Judge" had been published.7 The report described near-parity performance with cascade optimization.7 A Hacker News report on October 5 said decision models like Jev do not beat LLM-as-a-judge or traditional classifiers.8
Cloudflare's blog announced open-weight Clef and Clef-flash decision models on October 1.9 The blog said Clef tops the Jev Decision Index, with vision encoding, a 64k context window and superior latency versus competitors.10 Cloudflare also launched a reinforcement learning service for fine-tuning Clef on custom workloads.11 Latent Space reported on October 3 that Clef was post-trained from Qwen3.8-27B.12
Differing funding figures
Latent Space reported on September 25 that TypeSafe raised $1B+ at a $10B+ valuation, a week after a $200M round.13 TechCrunch reported on October 9 that TypeSafe AI raised $870 million at a $7.5 billion valuation, in a round led by Andreessen Horowitz.14 TechCrunch placed the round weeks after Jev's September 15 release went viral.14 The site typesafe.ai called it a Series A and said Martin Casado joins the board.15
Latent Space reported on October 10 that a tweet from Diogo Almeida said Sequoia leaked that TypeSafe passed $100M ARR in its first week.16
Latest reported
The Verge reported on October 15 that Microsoft launched Microsoft-Decision-1, which chooses the most efficient AI model for a task.17 According to The Verge, Microsoft claims the model beats others on speed and accuracy, and the launch came days after OpenAI's Decisions API.17 Zvi Mowshowitz wrote the same day that the model was reportedly a finetune of Qwen 3.5-9B, which the post found puzzling.18
Sources for the overview
- Simon Willison: TypeSafe AI unveils Jev decision model
- Latent Space: TypeSafe AI launches Jev System 1 model
- Simon Willison: Simon Willison releases llm-typesafe plugin
- Simon Willison: llm-typesafe 0.1a0 released
- Latent Space: TypeSafe Jev decision model release
- Nate B Jones: Jev launches on Vercel's AI Gateway
- Latent Space: Jev as Judge paper published
- Hacker News: Decision models like Jev don't beat LLM-as-a-judge or traditional classifiers
- blog.cloudflare.com: Cloudflare releases Clef decision models
- blog.cloudflare.com: Clef leads decision model benchmarks
- blog.cloudflare.com: Cloudflare launches RL fine-tuning service
- Latent Space: Cloudflare Clef decision model
- Latent Space: TypeSafe AI $1B+ fundraise at $10B valuation
- TechCrunch: TypeSafe AI raises $870 million at $7.5 billion valuation
- typesafe.ai: Typesafe AI raises $870 million Series A
- Latent Space: TypeSafe Jev hype and ARR claim
- The Verge: Microsoft launches Microsoft-Decision-1 model
- Zvi Mowshowitz: Microsoft-Decision-1 announced
The stories
- · TypeSafe Jev decision model launch
- · Jev as Judge paper published
- · Clef decision models and RL fine-tuning
- · Typesafe AI raises $870M
- · Microsoft-Decision-1 model
- · Gemini Decision-1 API launch