Braintrust says it's evals leader
Braintrust is touting itself as the evals platform used by Vercel, Replit, Notion and others and claims teams there run 12.8 eval experiments per day — a clear indicator of heavy, continuous compute demand. That scale of eval traffic creates predictable inference and benchmarking needs as they convert experiments into product features. (x.com)
Braintrust closed an $80 million Series B led by ICONIQ Growth on Feb. 17, 2026, at an $800 million post‑money valuation. (braintrust.dev)) The round included returning investors Andreessen Horowitz, Greylock, Elad Gil and Basecase Capital, according to the company announcement. (braintrust.dev)) CEO Ankur Goyal told podcasters that customers are running roughly 10× more evals year‑over‑year and that daily telemetry now dwarfs the total data logged in the product’s first year. (news.aakashg.com)) Braintrust says some advanced teams execute thousands of evals per day and that engineers can spend more than two hours each day iterating on experiments and scorers. (recapio.com)) To handle that volume the company built custom database infrastructure to ingest and query long agent traces, noting some agent sessions can generate hundreds of megabytes of trace data. (techfundingnews.com)) The platform centers on experiment‑tracking and real‑time tracing, offers on‑prem and private VPC deployment options, and ships SDKs for both TypeScript and Python integrations. (braintrust.dev)) Braintrust has published an open AI‑evals course and multiple GitHub example repos to accelerate adoption among engineering teams. (github.com)) The company says the Series B will fund expanded engineering and go‑to‑market hiring, new office openings, and continued development of observability features for production AI. (braintrust.dev))
