← Back to feed

TypeSafe AI has raised $870 million in a Series A round at a $7.5 billion valuation, coming just weeks after its Jev model went live on September 15. The round was led by Andreessen Horowitz, with Sequoia Capital joining alongside existing backer DCVC.

The fundraise is large even by 2026 standards and reflects an unusual velocity: a startup that closed a $40 million seed less than a month before its launch now carries a $7.5 billion price tag. TypeSafe says a third of Fortune 500 companies are already running Jev in production.

What Jev Actually Is

Jev is not a language model. Built on a transformer architecture, it produces calibrated probability scores across structured answer options rather than generating text. The output is a set of probabilities — what the company calls “calibrated decisions” — that software can act on directly without parsing or post-processing.

Supported task types: yes/no, multiple-choice, rating scales, and rubric-based grading of AI responses or agent actions. Each call returns a probability score for each option rather than a sentence.

TypeSafe co-founder Diogo Almeida, a former OpenAI researcher, frames it as a language mismatch problem: “We have been super good at human language for four years, but it’s not useful for automation because computers speak a different language.” The company positions Jev as the translation layer between AI reasoning and software execution.

The Team

Almeida co-founded TypeSafe in 2024 with Sasha Sheng, formerly a research engineer at Meta, and Erik Gafni, an engineer and entrepreneur. The founding team combines lab experience (OpenAI, Meta) with systems engineering depth.

Decision Models as a Category

TypeSafe is not alone. Microsoft shipped Decision-1 this week through Azure Foundry and OpenRouter, also targeting routing, classification, and prioritization workloads. Microsoft built Decision-1 on a post-trained Qwen3.5-9B base and claims 35x lower latency than GPT-6 Sol on structured decision tasks.

The category forming around these models — purpose-built classifiers and decision scorers rather than generative LLMs — reflects a push to cut token costs and latency on the subset of agentic tasks that never needed text generation in the first place.

Numbers

• Round size: $870M
• Valuation: $7.5B
• Investors: Andreessen Horowitz (lead), Sequoia, DCVC
• Prior raise: $40M seed
• Enterprise adoption (self-reported): ~33% of Fortune 500
• Time from launch to Series A close: approximately 3 weeks