Article
CommunityJev Media & Talks
Forbes: the $200M startup taking on AI's overconfidence
Forbes' Rashi Shrivastava profiles TypeSafe's $40 million DCVC-led seed at a $200 million valuation, and CEO Diogo Almeida's argument that software needs models reporting calibrated confidence rather than confident-sounding paragraphs.
- Format
- Article
- Reading time
- 3 min
- Published by
- Community
- Author
- Rashi Shrivastava
- Last verified
- Oct 2026
Highlights
- TypeSafe came out of stealth with a $40 million seed round led by deep tech VC DCVC, valuing the startup at $200 million, per a person familiar with the deal.
- Almeida: despite all the AI progress, the technology is still overconfident, with every answer presented like it is 100% accurate even when it is not.
- Almeida spent four years improving ChatGPT's responses at OpenAI, left in 2024, and started TypeSafe to train a new type of foundational model.
- The model returns numbers with probability estimates and scores meant to be ingested by software; outputs past a confidence threshold can be automated.
- Almeida claims that because the models are not generating text, they are 100x cheaper and faster to use.
Watch out
Copy reviewed via the publisher's search snippet on 2026-10-17; the page returned HTTP 403 to the fetcher, so this entry is grounded only on the captured excerpt and its note. The $200 million valuation is attributed to a person familiar with the deal, and the 100x claim is Almeida's.
Questions about this entry
Short answers drawn from the linked project or page.
The startup came out of stealth with a $40 million seed round led by deep tech VC DCVC, valuing it at $200 million according to a person familiar with the deal. The story is dated September 15, 2026, and was updated September 28.
AI's overconfidence, per the piece: every answer is presented like it is 100% accurate even when it is not. Almeida says doing real work in the enterprise requires AI to be honest about what it knows and how confident it is.
Numerical responses along with probability estimates and scores that indicate confidence in each metric, all meant to be ingested by software rather than read by a person, so businesses can tell which outputs are reliable.
If an output passes a certain confidence threshold it can be automated entirely, while more nuanced outputs can be flagged for human oversight. Almeida's framing: you can automate something instead of automating nothing because LLMs are too unreliable as an aggregate.
Asked a question like whether a property has a history of fires, the model would read and analyze the evidence and provide a percentage of how likely it is that there was a fire. Almeida says an LLM does not have the nuance to communicate that.
He spent four years at OpenAI improving ChatGPT's responses before leaving the firm in 2024. He says most intelligence is going to live inside software, but to automate any real work it has to be correct.
It is grounded on a search snippet because the page returned HTTP 403 to the fetcher. A note with the excerpt says an aggregator summary of the updated piece reports Vercel, Cloudflare and others adding Jev, and TypeSafe claiming GPT-5.6- and Sonnet 5-level workflow evals, all second-hand here.
Reactions & coverage
Posts, threads, and videos about this entry from around the web.
Reddit: LocalLLaMA debates whether TypeSafe's $870M raise is Jev hype.
More Jev resources like this
TechCrunch reports TypeSafe AI raised $870 million at a $7.5 billion valuation, led by Andreessen Horowitz with Sequoia and DCVC, weeks after Jev's September 15 release. The startup claims a third of Fortune 500 companies already use the model.
Jev Media & TalksArticle#community#press#funding
Community
Bloomberg Law reports TypeSafe AI raised about $870M at a $7.5B valuation in a round led by Andreessen Horowitz, weeks after the Jev launch video went viral. The startup says a third of Fortune 500 companies use Jev but declined to name any.
Jev Media & TalksArticle#community#press#funding
Community
SiliconANGLE reports TypeSafe closed $870M at a $7.5B valuation and explains the model's pitch: three request types, yes/no answers, picking from a list, or a score, return structured output and a confidence number instead of text.
Jev Media & TalksArticle#community#press#funding
Community
Andreessen Horowitz announces it is leading the investment in TypeSafe AI, calling Jev the fastest growing model it has seen at 1 trillion tokens generated in 3 days. The firm claims 1/100 to 1/500 frontier-model cost and 100x speed for classification, and says 25% of Fortune 500 enterprises have integrated Jev.
Jev Media & TalksArticle#community#press#funding
Community
TNW reports Jev reached nearly 13% of Vercel's paid teams within 24 hours, the AI Gateway's fastest adoption: twice the GPT-5.6 family's share and more than 6x Fable 5.1's, at $0.042 per million input tokens with output unmetered.
Jev Media & TalksArticle#community#press#adoption
Community
The Rundown's news brief reports TypeSafe's September 15 early access launch of Jev, a model that answers preset questions inside software, and flags that the company's can't-hallucinate guarantee concerns format, not correctness.
Jev Media & TalksArticle#community#press#calibration
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.