From the source
Mercor is running internal applications and open-model workloads on Vercel, using AI Gateway for model access and Passport for authentication, according to a Vercel case study.
- Mercor connects hundreds of thousands of domain experts with frontier AI labs and runs automated grading jobs, work that arrives in bursts and carries 24-hour deadlines.
- Its mix of direct integrations and third-party gateways could not supply enough token-per-minute capacity when demand surged, so requests hit provider limits and entered retry queues, stretching batch jobs beyond 12 hours.
- AI Gateway added inference capacity and failover routing between providers, and Mercor reports those jobs now finish in 2 to 3 hours.
- In its first 10 days on Gateway, the team routed 18 million requests and 63 billion tokens without hitting rate limits.
- Hundreds of engineers use open models through per-user keys with expiration dates, token limits and spend limits, so security can see which user, project and workload drives consumption.
- A few hundred people across business teams now build on Vercel, where the project count grew from fewer than a dozen to roughly 2,000, and Passport handles sign-in through an existing identity provider.
