Orchestra launched in June as Understudy Labs. The new name reflects what the product became, a single channel for inference that moves repeated work to cheaper models once they prove they can do it.
The gap the product filled.
The founders' view is that the fastest way to grow is to serve buyers who already pay for a solution that still leaves a gap. They found that gap in inference, a market they put at $104B. Working with AI teams in big tech and fast-growing AI-native startups showed that routine tasks rarely need the most capable model. As they put it, "you don’t need a PhD to work in the mailroom."
The product began as a simple pass-through gateway. By the end of the company's Y Combinator cohort, it was an enterprise product, shaped first with early design partners and then scaled with larger enterprise partners.
How it works.
Teams swap a single URL to send their inference through Orchestra. Orchestra uses the resulting traces to build evals in the background. Those evals compare routes on the same work, showing which tasks need frontier capability and which suit a specialist. When a post-trained specialist beats the frontier model on a team's workload, it gets promoted and the bill goes down.
Orchestra does not pre-train models and does not rely on forward-deployed engineers. It builds and routes each task to the most efficient model.
Why the name changed.
"Understudy" cast the company as a backup to the frontier labs. Customers showed they would move all of their inference spend through one channel once performance was proven, so the founders wanted a name that matched that role. They also call inference the fastest-growing market in history, and wanted a brand that fit that ambition.
The full announcement covers more of the company's first months. If inference is a growing line in your budget, evaluate your workflows.