Baseten Acquires Parsed to Enable Companies to Own Their Intelligence

The acquisition adds world-class reinforcement learning and post-training expertise to deliver superior inference quality and performance for Baseten customers via specialized intelligence

SAN FRANCISCO--(BUSINESS WIRE)--Baseten, the company powering inference for the world’s fastest-growing AI products, today announced that it has acquired Parsed, a reinforcement learning startup specializing in post-training and continual learning for large models. The Parsed team, a group of preeminent researchers and engineers with deep expertise in reinforcement learning, has already been working closely with Baseten customers as a partner. With this acquisition, the Parsed team and technology will become part of Baseten to unify inference, data, evaluation, and post-training in a single platform.

“Inference is the stickiest and most important part of the AI stack,” said Tuhin Srivastava, co-founder and CEO of Baseten. “What makes inference truly valuable over time is continual learning: using real production data and evaluations to train better, faster, cheaper models. Parsed is world-class at reinforcement learning and post-training, and bringing them in-house lets us cover the entire AI lifecycle.”

From medical scribing to AI-powered go-to-market tools, companies are increasingly moving from general-purpose models toward specialized models optimized for a single, high-value task. By uniting applied AI research, flexible infrastructure, and seamless developer tooling, Baseten enables companies operating at the frontier of AI to scale custom and open-source models to millions of users in production. Parsed complements this foundation with a vertically integrated approach to post-training, working with customers to design the right learning signals from real-world usage, defining what should be rewarded, what should be penalized, and how that feedback shapes gradient updates.

“We believe that the world doesn’t need ever-bigger general-purpose models, it needs models that deeply understand a specific job and keep getting better at it,” said Mudith Jayasekara, Cofounder and CEO of Parsed. “Baseten is the best place in the world to run those models in production, and we’re excited at what's possible when we bring inference and training together.”

The Parsed team offers a rare combination of deep reinforcement learning research expertise and a strong, practical focus on solving customer problems. The three cofounders started the company after meeting at Oxford whilst pursuing their PhDs: Chief Scientist Charles O’Neill, and Rhodes Scholars Mudith Jayasekara and Max Kirkby. The team brings a wealth of prior ML research experience with ML Alignment & Theory Scholars, Cambridge, Stanford, NASA, and Johns Hopkins.

“Ultimately, customers don’t buy ‘training’ or ‘inference’ in isolation; they buy AI that does real work,” added Srivastava. “By combining Parsed’s training systems with Baseten’s inference platform, we’re building the definitive place to run specialized, high-performance models that keep getting better the more you use them.”

The news follows a series of notable developments for Baseten. It recently announced the close of a $150 million Series D fundraise, which valued the company at $2.15 billion and brought its total capital raised to over $285 million.

About Baseten

Baseten is the leader in inference for high-scale AI products, providing one of the industry’s most advanced inference stacks. Purpose-built for performance, reliability, and cost-efficiency, Baseten enables teams to build and scale the next generation of AI products. Through applied research, production-grade infrastructure, and a seamless developer experience, customers can infinitely scale open-source, custom, and fine-tuned models in production. Backed by BOND, CapitalG, IVP, Spark, Greylock, Conviction, and others, Baseten is trusted by the fastest-growing AI companies to power their most ambitious products. Learn more at www.baseten.co.