VendorsMSFT +2.38%OCT 11, 2026

TypeSafe AI's Jev hits $7.5B valuation at $0.042 per million tokens; Microsoft matches price same day

Andreessen Horowitz-led round values the decision-model startup 24 days after launch as Microsoft Decision-1 lists at the identical $0.042/M input price.

BY ELIN BJORKLUND · VENDOR ANALYSIS

TypeSafe AI closed an $870 million round led by Andreessen Horowitz on October 9, 2026, valuing the two-year-old startup at $7.5 billion just 24 days after shipping Jev, its “decision model” that scores and classifies rather than writes. The same day, Microsoft listed Decision-1 on Foundry at $0.042 per million input tokens, the identical price TypeSafe charges for Jev 1.13.

That price convergence, down to the tenth of a cent, is the actual story.

TypeSafe was co-founded in 2024 by Diogo Almeida, a former OpenAI researcher who worked on InstructGPT, ChatGPT and GPT-4, and Sasha Sheng, previously of Meta. The seed, priced over a year before launch, was $40 million at a $200 million valuation. The Series A marks a 37.5x step-up for roughly 11.6% of the company, which still has just over 20 employees. Almeida told the WSJ Jev was generating a trillion tokens a day by the end of September; a16z pegged the first trillion at three days post-launch. On Vercel’s AI Gateway, nearly 13% of paid teams tried Jev inside 24 hours, more than twice the reach of any prior model launch.

The hyperscaler response arrived almost instantly. AWS shipped Strands Decider 2B on October 1. OpenAI, which four weeks earlier had no Decisions API on its roadmap, released one on October 6 at $0.10 per million tokens. Cloudflare’s Clef-flash was listed at $0.090 on October 3 and cut to $0.038 by week’s end; the full Clef model runs $0.240. Microsoft’s Decision-1, built on Qwen3.5-9B, posted a Foundry-ranked 36-test, nearly 150,000-question benchmark showing 1.3% answer drift across eight perturbation types and zero flips when options were paraphrased, reversed, or shuffled. Stanford and Nvidia’s open CLM-8B runs up to nine times faster than Jev on some tasks.

The customer math is unforgiving. The talent marketplace Jack & Jill moved candidate screening from Gemini 3.1 Flash Lite at $0.755 per thousand candidates to Jev at $0.092, an 88% cut with a 90% confidence interval of −0.4 to +1.8 points on winner retention. Median screening time halved, from 20.3 seconds to 10.3. Jack & Jill now runs Jev in more than 15 workflows and would need 8.2x usage growth before its AI bill returns to what it was.

TypeSafe claims a third of the Fortune 500; a16z says a quarter. The gap is the pitch deck.

Sources

About the author

Elin Bjorklund

VENDOR ANALYSIS

Elin Bjorklund follows the application-layer AI vendors — productivity, design, creative — with a focus on revenue line breakouts and run-rate disclosures. She does deep reads on vendor financials when companies break out AI-attributable revenue.