Anthropic quietly shipped two significant API upgrades that expand what developers can build with Claude. The max_tokens cap on the Message Batches API has been raised to 300,000 tokens for Claude Opus 4.6 and Sonnet 4.6, enabled via the output-300k-2026-03-24 beta header. This is a game-changer for use cases requiring long-form output — large code generation tasks, structured data extraction, and comprehensive documentation can now be produced in a single API call.
The second upgrade is equally impactful: API code execution is now free when used alongside web search or web fetch tools. Sandboxed code execution allows Claude to run Python code during a conversation to process data, generate visualisations, or validate its own outputs. By making this free when combined with web tools, Anthropic is encouraging developers to build agentic workflows where Claude can search the web, fetch content, and immediately process it with code — all within a single API call at no additional cost.
Both features reflect Anthropic's strategy of removing friction from agentic development. The 300K output limit addresses a long-standing complaint from developers building code generation pipelines: previous output limits forced artificial chunking of large codebases. The free code execution removes a cost barrier that discouraged developers from letting Claude verify its own work — a pattern the leaked source code revealed is central to Claude Code's internal architecture.
For context engineers building production systems, these updates mean more ambitious single-pass code generation, cheaper agentic pipelines that combine search and computation, and fewer workarounds for output length limitations. The Batch API's 300K limit is particularly relevant for teams using Claude to generate or refactor entire modules.
