Firecrawl has released version 2.5 of its web data API, focusing on extraction quality and coverage for developers building AI agents and data pipelines. The update centers on a custom browser stack and a Semantic Index designed to return cleaner, more reliable page data without forcing customers to rewrite integrations.

A custom browser stack for data quality

Firecrawl v2.5 uses a custom-built browser stack that detects how each page renders, whether as a PDF, paginated table, or dynamic JavaScript application, then converts that content into clean, AI-ready formats. By indexing complete pages rather than partial snippets, the company says the system preserves structure and outperforms narrower scrapers on quality benchmarks.

Semantic Index for faster, more reliable access

At the center of the release is the Semantic Index, which Firecrawl says already serves about 40% of API calls. It stores full-page snapshots, embeddings, and structural metadata so users can retrieve data as of now or from the last known good copy. A maxAge parameter lets developers set freshness requirements for each request.

Availability

Firecrawl v2.5 is available immediately and requires no code changes for existing users. The company also plans to open-source its web data retrieval benchmarks in the coming weeks. More information is available at firecrawl.dev.

Decoded Take

Web data has become a bottleneck for agent systems that need reliable page state, not just brittle scrapes. Firecrawl’s bet is that a managed browser plus semantic cache can become the default retrieval layer for AI apps the way search APIs once became default for content discovery. Open-sourcing benchmarks will matter: if independent tests back the quality claims, Firecrawl strengthens its position as infrastructure. If not, rivals will keep competing on coverage, anti-bot resilience, and price.