New Claude Cookbook guides cover search benchmarks, safety fallbacks, and multi-agent patterns
Anthropic added several technical guides to the Claude Cookbook in June 2026. Mengting Li’s walkthrough shows how to build a Messages API harness that reproduces the published DeepSearchQA and BrowseComp scores using programmatic tool calls, server-side compaction, and task budgets.
A second guide breaks down what happens when the safety classifier fires on Claude Fable 5 and how to fall back to Opus 4.8, either server-side or client-side through the SDK, accounting for streaming behavior and the latest billing changes.
A third guide from Paul Chen reduces two async multi-agent patterns to their basic mechanics: a fixed team of N agents passing messages through a shared hub, and dynamically spawned async subagents.
Up next
A zoom tool pushes chart-reading accuracy to 73%, Video
Give Claude a zoom tool and Fable 5 jumps from 29% to 73% on Chartography, a 100-question benchmark of dense real-data charts. Sonnet 5 goes from 13% to 44%. The model requests a region and gets that slice back from the original at full resolution.
Related stories
- Claude Managed Agents can now hand tasks off to each other
- Claude helped make claude.ai 3x faster in two weeks
- Claude system prompts documentation updated
- Claude threads shrink long requests from 800 KB to 1.8 KB
- Claude API enforces thinking blocks and moves to Rust
- Claude API raises the request cap from 32 MB to 128 MB
Comments
No comments yet. Be the first.
Join the conversation
Sign in to leave a comment. It appears after moderation.