Model

Mercor names chain-of-thought author Jason Wei chief scientist

The researcher behind chain-of-thought prompting and three benchmarks the industry grades itself with joins a $20 billion expert network, one week after its $750 million Series D, to lead a push selling lab-grade AI evaluations to every company.

Vincent JiangVincent Jiangthree-minute read
Share
Headshot portrait of Brendan Foody, co-founder and chief executive of Mercor, against a grey background
Brendan Foody, Mercor's co-founder and chief executive, hired Jason Wei as the company's chief scientist a week after its $750 million Series D.

The chain-of-thought author leaves Meta's lab

Jason Wei, whose 2022 chain-of-thought paper showed that models reason better when they write out their steps, is joining Mercor~$2B — Mercor, private, latest valuation ~$2B as chief scientist 192. Mercor said so on November 18 19, and Wei confirmed it in his own words: "After an amazing year at Meta's−0.60% — Meta, down 0.60 percent today TBD lab, I am very excited to join Mercor as Chief Scientist" 2.

He made his name at Google Brain, moved to OpenAI$1.18T — OpenAI, private, latest valuation $1.18T in 2023, and created SimpleQA, BrowseComp and HealthBench, three of the tests the industry now grades itself with 19. His new employer, founded in 2023 by Brendan Foody, Adarsh Hiremath and Surya Midha, pays a network of more than five million specialists to train and grade AI models 310.

One week after the money arrived

The hire lands seven days after a $750 million Series D valued Mercor at $20 billion, with Nvidia−2.79% — Nvidia, down 2.79 percent today, Salesforce−2.73% — Salesforce, down 2.73 percent today and Zoom+2.02% — Zoom, up 2.02 percent today among corporate backers and an employee tender folded in, so an undisclosed slice is staff cash-out rather than new capital 37. The Inference covered the round and its tasker pay floor on November 11.

Foody, Mercor's co-founder and chief executive, made the market case onstage at the Cerebral Valley summit: agents now need top engineers, lawyers, doctors and bankers to design evaluations and practice environments, and the ceiling is the set of tasks humans still do better than models 6. Wei's line matches: "Data will shape the next frontier as more companies train AI models than ever before" 2.

Mercor's valuation went from an implied $2B to $20B in three rounds

Price doubled twice in under two years; the latest round landed a week before the Wei hire

  • Valuation
  • Estimate
$0B$5B$10B$15B$20BSeries B (2025, implied)Series C (Oct 2025)Series D (Nov 2026)Round included an employee tender
Data
Valuation
Series B (2025, implied) (estimate)$2B
Series C (Oct 2025)$10B
Series D (Nov 2026)$20B
Series B implied from Mercor's '5x' Series C claim (FinanceInTime computation); Series C and D from company announcements and round coverage. US dollars, billions.3,7,8

A research stack already assembled

Wei inherits machinery, not a blank page. The research team, announced on October 8, already includes Edward J 4. Hu, first author of LoRA, the standard recipe for cheaply fine-tuning large models 4. Mercor's APEX benchmarks score AI on consulting, banking, corporate law, accounting and software work, and a published training guide lifted Qwen3.5-397B from 16% to 27% on APEX-Agents 4.

In July, Mercor bought Deeptune, whose software recreates enterprise applications for agents to practice in, which had raised a $43 million Series A led by Andreessen Horowitz 5. Annualized revenue passed $2 billion in June, on Foody's own figure 7. The stated aim is more training signal from each expert-hour, because expert time is expensive and limited 4.

The customers are the competition

Frontier labs remain the main buyers of expert training data, says Jai Das of Sapphire Ventures, one of the round's investors 7. Mercor's announcements name no enterprise customer and no split of revenue between labs and everyone else 38. Every lab that buys Mercor's experts is also a lab that could hire its researchers. The next reading that matters is Wei's first Mercor benchmark: adopted by companies, it turns an expert-hours marketplace into a research company; ignored, the $20 billion price was set for a supplier 38.

Get Model by email, free

The AI labs, their models, and the business behind them. In our daily email, every weekday.

Loading META

Reader comments

0 comments

    Model

    HeyGen's Avatar 6 scored 84% on realism, but its 42.2% pick rate was below chance

    Avatar 6 launched November 17, 2026, with an 84% realism figure and first place in a blind arena; the score doubles a below-chance pick rate, and the arena capped 97% of clips at 35 seconds. That is the evidence base under a company claiming $200 million of annual revenue.

    Model

    MiniMax hit a record low, about 90% below its January peak, as a US rule gates its rented Nvidia compute

    Washington's new remote-access rule puts a US license between China's listed model makers and the offshore Nvidia capacity they rent. MiniMax, the purest listed bet on Chinese foundation models, holds the price tag.

    Model

    OpenAI cuts flagship pricing 20% to match Anthropic as heavy users call GPT-6.2 Astra unusable

    OpenAI cut its flagship's price to match Claude Fable 5.2's rate, and the users who run models hardest rejected it, 24 hours before Anthropic priced a $1.4 trillion IPO on the assumption the quality gap holds. The independent labs say the gap is real, for now.

    Model

    Microsoft books Azure revenue on Chinese open models running on Baseten hardware it does not own

    Microsoft now bills customers' committed Azure dollars for open models from Chinese labs running on Baseten's hardware, a toll booth earning rent on compute it does not own.

    Model

    Alibaba stops giving away its best AI model, three days before earnings

    Alibaba launched Qwen4-Max on November 16 and kept the 2.4-trillion-parameter flagship behind its own paid API, promising open weights only for smaller models. Thursday's report is the first test of whether Qwen's three billion downloads become cloud revenue, and it lands two days after a US rule put the chips behind the model under license.

    Model

    Perplexity claims $1 billion of revenue on five times less spend than Anthropic's $1.4 trillion IPO

    Perplexity says it hit $1 billion in recurring revenue on a fraction of Anthropic's spend, three days before Anthropic prices a $1.4 trillion IPO on the opposite assumption. Whose economics hold will only be settled by a filing Srinivas now says could come in early 2027.

    Subscribe to TheInference for free