Article
CommunityJev Media & Talks
The New Stack: 'totally useless for computers'
Adrian Bridgwater's September 21 report on TypeSafe's launch pitch: a model that does not write but decides, a new architecture and RLCD training, and Vercel's Jev adoption numbers through its AI Gateway.
- Format
- Article
- Reading time
- 3 min
- Published by
- Community
- Author
- Adrian Bridgwater
- Last verified
- Oct 2026
Works forJev for Decision gating
Used inJev for Software & DevTools
Highlights
- Quotes Diogo Almeida's September 15 post: LLMs generate sequentially, which is 'great for a natural conversation, but totally useless for computers.'
- TypeSafe emerged from two years in stealth with a $40M seed led by DCVC; the report describes a new architecture, a new sampler, and RLCD training.
- TypeSafe's reported claims: 20-200x faster, 40-400x cheaper with output tokens free, and 'can't hallucinate' because outputs are typed decisions.
- Vercel figures cited: nearly 13% of paid teams used Jev within 24 hours, 2x the GPT-5.6 family and more than 6x Fable 5.1's share.
- Cites Flavio Copes' write-up calling Jev 'a smart if statement' and a small component inside an ordinary application.
Watch out
Copy reviewed against the article text captured on 2026-10-17; the page is client-rendered, so the full text was recovered through the search index rather than fetched HTML. The speed, cost, and hallucination claims are TypeSafe's own, reported second-hand, and the adoption figures come from Vercel.
Questions about this entry
Short answers drawn from the linked project or page.
Almeida's: LLMs generate text sequentially, which suits conversation but serves computers poorly. TypeSafe built Jev, a System One model, so software can get typed decisions with calibrated probabilities and set its own thresholds for autonomous action versus human review.
TypeSafe's launch claims as reported: 20-200x faster and 40-400x cheaper than comparable LLMs on classification work, with output tokens free. The article relays the 'can't hallucinate' framing through typed outputs, and it presents these as the company's claims.
Per Vercel figures cited in the piece, Jev reached more than twice as many paid teams in its first 24 hours on the AI Gateway as any previous model launch: nearly 13% of paid teams, 2x the GPT-5.6 family and more than 6x Fable 5.1's share.
The piece describes a new architecture, a new sampler, and a training algorithm TypeSafe calls Reinforcement Learning for Calibrated Decisions. It traces the motivation to Almeida's critique of RLHF chat models, including mode dropping and overconfidence.
Flavio Copes' independent deep dive, cited by The New Stack: Jev is not a chatbot or a coding model but a small component inside a regular application, added where code needs a single judgment, with most calls around 100ms at $0.042 per million input tokens.
After the 19th-century economist William Stanley Jevons and the Jevons Paradox: as technology makes a resource more efficient, total consumption of it rises rather than falls. The piece ties the name to TypeSafe's intelligence-per-dollar framing.
It had emerged from two years in stealth with a $40 million seed round led by DCVC, weeks before the $870M round at a $7.5B valuation that later coverage in this batch reports. Almeida's closing note in the piece: 'building prod, not God.'
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Hacker News: Simon Willison's overview of Noul, Choice, and Score, and where a decision model fits.
benwerd53 pointsJev introduces a new shape of LLMRead the thread on Hacker NewsHacker News: swyx posts the Latent Space interview with Diogo Almeida.
swyx4 pointsJev: System One Models for Prod, Not God – With Diogo Almeida, CEO, TypeSafe AIRead the thread on Hacker NewsX: Diogo Almeida in five minutes
X: LLMs vs. Jev, clearly explained
More Jev resources like this
LangChain's September 25 post argues Jev's typed decisions fit LangGraph's deterministic orchestration, with a litigation review example and a Stagehand browser case where act() latency fell to 0.46 seconds.
Jev Media & TalksArticle#community#press#explainer
Community
A builder's walkthrough of System One, a dual-process browser agent that resolves routine steps with four parallel Jev questions in 70-150ms and falls back to Gemini or Claude below 0.55 confidence.
Jev Media & TalksArticle#community#press#explainer
Community
Tom Joseph's analysis walks through Jev's typed primitives and finds three claims that survive scrutiny: the parallel sampler, the price and speed, and the RLCD training whose calibration plot is unpublished.
Jev Media & TalksArticle#community#press#explainer
Community
A playground operator's audit of every published Jev number: TypeSafe's four-workflow evals, the favourable homepage demo, independent tests and its own calls. It concludes the price and speed are real and the intelligence claim is roughly mid-tier frontier on decision workflows.
Jev Media & TalksArticle#community#press#explainer
Community
AICrier's news-feed summary of Celeris-1 Decision: a multimodal diffusion model for bounded structured decisions that Celeris says reaches 80% on JevBench, 7 points above Jev 1.13.0, with 67 ms median latency.
Jev Media & TalksArticle#community#press#competition
Community
YourStory's introduction to Jev, the transformer-based model that returns structured decisions with probabilities instead of paragraphs or code, covering the September 15 stealth exit, the $40M DCVC-led seed, and TypeSafe's pricing claims.
Jev Media & TalksArticle#community#press#explainer
Community
From the Jev guides
Original write-ups on what this entry is for.
From the Jev AI model community
Posts from builders shipping with Jev right now.