TypeSafe raises $870 million at $7.5 billion valuation as Jev bets automation does not need text

TechCrunch reports that TypeSafe AI, the startup behind the Jev model, has raised $870 million at a $7.5 billion valuation. Andreessen Horowitz led the round, with participation from Sequoia and existing investor DCVC. Jev launched on September 15 and went viral almost immediately (developers thrilled).

TypeSafe was co-founded in 2024 by Diogo Almeida, previously a researcher at OpenAI; former Meta research engineer Sasha Sheng; and engineer and entrepreneur Erik Gafni. Almeida’s pitch is that computers speak a different language than people, and that automation should be built around that.

Jev is built on a transformer architecture but is not a large language model and does not output text. It produces probabilities, which the company calls calibrated decisions. TypeSafe claims Jev is significantly faster and uses far fewer tokens than LLMs, and pitches it for automating tasks rather than writing prose or code. Our explainer compares it with Fastino GLiDE, GLiNER2.5-Decide and open-source rivals (decision models explained), and a coding guide covers typed decisions and calibrated confidence (Jev coding guide).

The approach targets a large, quiet slice of enterprise AI: yes-or-no, pick-one and route-this decisions made millions of times a day. Many companies today ask a language model to make such a call, then parse a label out of its prose, which is slow, expensive and brittle. A model that answers directly with a probability fits automation more cleanly, and an explicit confidence score invites a threshold and a human check in a way a fluent paragraph does not. If Jev’s outputs are genuinely calibrated, that is a safety property as well as a speed one.

The adoption figure is TypeSafe’s own claim, and “using” can range from a pilot to production. Calibration is easier to promise than to prove across messy real-world data, and TechCrunch’s report does not include independent evidence either way. A $7.5 billion valuation weeks after launch prices in substantial success before the market has tested it.

Competition is arriving fast. OpenAI has put its own typed-answer Decisions endpoint into public beta, and coverage has framed it as a Jev clone that could help control swarming agents (OpenAI’s clone). Amazon has released its own version (Amazon clone), and Contrastive-LM claims its open CLM-8B scores agent actions up to nine times faster than Jev (CLM-8B). TypeSafe’s edge will likely rest on price, speed and trust rather than novelty.

Faster, cheaper decisions mean more decisions made without a person involved, in claims, moderation, routing and screening. That is useful for low-stakes calls with sensible thresholds and harmful when a probability quietly becomes a denial. How customers set and disclose thresholds will matter as these models spread. More in our funding coverage, startups coverage and enterprise coverage.