Structured Outputs on the Claude Developer Platform (API)

Claude adds “structured replies” — devs split between relief and eye‑roll

TLDR: Anthropic launched structured outputs for Claude, promising responses that fit your requested format. The community’s split: relief that apps won’t break, eye‑rolls that it’s late, and curiosity about whether it’s just fancy retries—still a big deal for reliable AI agents and fewer frustrating bugs.

Anthropic just gave Claude a chore chart: structured outputs in public beta for Sonnet 4.5 and Opus 4.1. Translation for non‑devs: the AI now promises to answer in a tidy, predictable format (like a filled‑out form), so apps don’t crash because the robot forgot a comma. Use it with JSON (a data recipe) or “tools” (pre‑defined actions), and your output should always fit the shape you asked for. OpenRouter’s COO is hyped, saying agents finally run smooth every time. But the crowd? Spicy.

The hottest take comes from jascha_eng, who basically roasted Anthropic: “it was crazy they’re only adding it now”. They’ve been chaining multiple AI “agents” together and needed clean hand‑offs ages ago. On the flip side, barefootford casually confessed they went unstructured on Claude and “haven’t had any issues (yet)”, but admits guarantees are comforting. Meanwhile, jmathai dredged up old lore—remember the hack where you had to paste the start of your expected JSON like {"data": into your prompt? “I couldn’t believe that was an actual recommendation.” The meme of the day: no more “feral JSON.”

And the nerdy speculation is sizzling: mkagenius wonders if this is just clever retries behind the curtain or true model smarts. Either way, multi‑agent setups, data extraction, and complex searches should breathe easier. Haiku 4.5 support is “coming soon.” Docs are here: Anthropic Docs.

Key Points

  • Structured outputs are available in public beta on the Claude Developer Platform for Sonnet 4.5 and Opus 4.1.
  • The feature enforces API responses to match developer-specified JSON schemas or tool definitions.
  • Benefits include eliminating parsing errors, reducing failed tool calls, and improving reliability in production applications and agent workflows.
  • Structured outputs can be used via JSON schemas in requests or via predefined tool specifications.
  • Support for Haiku 4.5 is coming soon; documentation covers schema types, examples, and best practices.

Hottest takes

“it was crazy that anthropic is only adding it now” — jascha_eng
“I couldn’t believe that was an actual recommendation” — jmathai
“is it just retries while generating tokens” — mkagenius
Made with <3 by @siedrix and @shesho from CDMX. Powered by Forge&Hive.