Browserbase Changelog
Browserbase shifted toward embedding browser automation deeper into agent workflows. Model Gateway abstracted away model provider friction for Stagehand, letting developers switch between OpenAI, Anthropic, and Gemini with a one-line configuration change. Browserbase Functions shipped execution directly on their infrastructure—agents now deploy alongside browser sessions, cutting latency by up to 70% and eliminating orchestration overhead. On the data side, Fetch API arrived as a lightweight alternative to full browser sessions (~$1 per 1,000 pages), and Browserbase Search enabled agents to query the web in a single call, each grounded in the platform's existing infrastructure rather than external dependencies.
March expanded Browserbase's core capabilities for agent developers. The Fetch API arrived for lightweight content retrieval without browser overhead, while the new Search API powered by Exa lets agents query the web before taking action. Infrastructure improvements landed too: the MCP server migrated to Browserbase-managed endpoints for better reliability, and free plan concurrency tripled to 3 browsers. A partnership with Prime Intellect made browser agent training more accessible through scalable Browserbase environments and the BrowserEnv reinforcement learning framework.