🧭 Structured Outputs — Generally Available Across the Claude API

Anthropic has moved Structured Outputs to general availability across the Claude API, graduating the feature from its earlier beta status. Structured Outputs guarantees that Claude will return a response that is valid JSON conforming exactly to a developer-specified schema — eliminating the parsing failures, schema violations, and hallucinated field names that previously required defensive wrapper code in any application that needed reliable JSON from a language model.

How Structured Outputs works

Developers pass a JSON Schema alongside their request; Claude's output is guaranteed to match the schema or the API returns an error rather than an invalid response. This is enforced at the model level through constrained decoding, not post-hoc validation.

import anthropic

client = anthropic.Anthropic()

response = client.messages.create(
    model="claude-sonnet-4-5",
    max_tokens=1024,
    messages=[{
        "role": "user",
        "content": "Extract the key details from this invoice..."
    }],
    response_format={
        "type": "json_schema",
        "json_schema": {
            "name": "invoice",
            "schema": {
                "type": "object",
                "properties": {
                    "vendor": {"type": "string"},
                    "amount": {"type": "number"},
                    "due_date": {"type": "string"}
                },
                "required": ["vendor", "amount", "due_date"]
            }
        }
    }
)

What reaches GA

• JSON Schema support — full JSON Schema draft 2020-12 support including nested objects, arrays, enums, and optional fields; anyOf and oneOf are supported for polymorphic schemas
• Streaming-compatible — Structured Outputs works with streaming responses; partial JSON is streamed token by token and validated on completion
• All Claude 4.x models — available on Haiku 4.5, Sonnet 4.5, and Opus 4.5; not available on legacy 3.x models
• No beta flag required — the feature is now in the default API surface; existing beta users do not need to make any changes beyond removing the beta header from their requests

🧭 Claude Code for JetBrains IDEs — Public Beta Now Available

Anthropic has launched a public beta of Claude Code for JetBrains, an IDE plugin that integrates Claude Code's agentic coding capabilities directly into IntelliJ IDEA, PyCharm, WebStorm, GoLand, and other JetBrains IDEs. The plugin extends Claude Code's CLI capabilities with direct IDE integration: Claude can read and write files, navigate the project structure, and see the active editor context without requiring the developer to copy-paste code into a separate terminal session.

Features in the beta

• Embedded terminal panel — a dedicated Claude Code panel in the IDE runs the CLI directly, with syntax-highlighted output and clickable file references that open in the editor
• Active file context — Claude Code automatically receives the path and content of the file currently open in the editor as implicit context, reducing the need to specify file paths manually
• Diff viewer integration — proposed file changes appear in JetBrains' native diff viewer rather than the terminal, with accept/reject buttons that apply changes directly to the project
• Project-level CLAUDE.md reading — the plugin reads CLAUDE.md from the project root and relevant subdirectories, providing Claude Code with project context in the same way the standalone CLI does

The beta is available as a JetBrains Marketplace plugin. A VS Code extension released earlier in 2025 has been downloaded over 300,000 times, and Anthropic anticipates strong JetBrains uptake given the platform's large enterprise Java and Kotlin developer base.