Mercor
An AI recruitment and evaluation platform that uses AI for talent matching and assessment.
Timeline (last 90 days)
Training environments and evaluation benchmarks use the same technical components but for different purposes: a score in training is a reward (the criterion being whether it can make the model stronger), while a score in evaluation is a measurement (the criterion being reliability, validity, and the ability to predict real post-deployment performance).
Reliability is a prerequisite for evaluation validity; evaluation should be treated as a product requiring long-term maintenance rather than a one-off ad hoc tool.
Zhang Yonghao divides the evolution of the labor market into three stages: 1.0 talent recruitment (such as BOSS Zhipin and Liepin), 2.0 flexible employment (such as Mercor), and 3.0 the Agent labor era.
This round was led by FirstMark, with founders of companies including Stripe, Perplexity, Datadog, and Mercor participating as angel investors.
Mercor's labor index research shows that 90% of US work hours are spent on tasks not covered by current AI benchmarks.
After Mercor migrated to Vercel AI Gateway, engineers accessed hundreds of models through a single expirable key and could set budgets; in the first 10 days it processed 63 billion tokens without hitting any rate limits.
All Facts (6)
Training environments and evaluation benchmarks use the same technical components but for different purposes: a score in training is a reward (the criterion being whether it can make the model stronger), while a score in evaluation is a measurement (the criterion being reliability, validity, and the ability to predict real post-deployment performance).
50%UnverifiedReliability is a prerequisite for evaluation validity; evaluation should be treated as a product requiring long-term maintenance rather than a one-off ad hoc tool.
50%UnverifiedZhang Yonghao divides the evolution of the labor market into three stages: 1.0 talent recruitment (such as BOSS Zhipin and Liepin), 2.0 flexible employment (such as Mercor), and 3.0 the Agent labor era.
50%UnverifiedThis round was led by FirstMark, with founders of companies including Stripe, Perplexity, Datadog, and Mercor participating as angel investors.
50%UnverifiedMercor's labor index research shows that 90% of US work hours are spent on tasks not covered by current AI benchmarks.
50%UnverifiedAfter Mercor migrated to Vercel AI Gateway, engineers accessed hundreds of models through a single expirable key and could set budgets; in the first 10 days it processed 63 billion tokens without hitting any rate limits.
50%