TechCrunch and trade press cover Jev as a non-LLM decision model
On 18 September 2026, TechCrunch published a feature on TypeSafe AI’s Jev: a transformer that returns calibrated decisions, not language. Founder Diogo Almeida (ex-OpenAI, RLHF lineage) told TechCrunch the problem is optimizing for human language when computers need a different interface.
The same week, InfoWorld (17 Sep) and The Register (16 Sep) covered the same shape: state in; typed Choice, Score, or Boolean/Noul out, with probabilities. The Register cites TypeSafe’s ~$40M raise. Some later coverage repeats higher valuation figures that ModelSystem.One has not independently verified.
Quotes in the TechCrunch piece (Vercel classifier swap, Bryo AI email test, Earendil on confidence thresholds) are third-party anecdotes from that article — not ModelSystem.One benchmarks.
Vendor speed and cost multiples remain not independently verified by ModelSystem.One.