Claude Platform

Compliance API chat endpoints now also return chats from the unified Claude experience, in beta for Claude Enterprise organizations, using the existing Compliance Access Key.

Prompt cache reads on Claude Sonnet 5.5 drop from $0.20 to $0.10 USD per million tokens, now 0.05x the base input price instead of 0.1x. Cache writes and all other prices are unchanged.

The web_fetch tool in Claude Managed Agents now fetches only URLs that have already appeared in the session, via user messages, web_search results, or previously fetched pages; URLs appearing only in Claude's output, system prompt, attached documents, or tool outputs return a url_not_in_prior_context error. The restriction is intended to reduce data exfiltration risk.

Claude Haiku 5.5 (claude-haiku-5-5) is tuned for high-volume, latency-sensitive work with a 1M token context window, 128k max output tokens, and adaptive thinking with an effort parameter. It is available on Claude API, Amazon Bedrock, Claude Platform on AWS, Google Cloud, and Microsoft Foundry.

In Claude Managed Agents, cloud environments with limited networking now apply allowed_hosts to the web_search and web_fetch tools: web_fetch returns a url_not_allowed error for non-matching hosts, web_search omits their results, and neither tool returns results when allowed_hosts lists no hosts.

Python and TypeScript SDKs now include beta classes for the browser use and computer use tools; subclass one and write a method per tool against your own browser or desktop automation, while the SDK runs the tool loop, URL and file policies, and approval callback.

Claude Max and Team plans now include monthly API credits, claimable through the API credits for subscribers feature.

Added capabilities.server_tools to the Models API, so GET /v1/models and GET /v1/models/{model_id} now report whether each model accepts web search and code execution tools. capabilities.server_tools.code_execution indicates code execution tool support.

Added capabilities.thinking.types.disabled to the Models API, so GET /v1/models and GET /v1/models/{model_id} now report whether each model accepts thinking: {type: "disabled"}.

Added a line field to the Models API. GET /v1/models and GET /v1/models/{model_id} now return the model line each model belongs to (e.g., Claude Opus 4.5 and Claude Opus 4.6 both report opus). Use line to group models without parsing their IDs. line is null for…

Claude Sonnet 4.5 (claude-sonnet-4-5-20250929) is deprecated with retirement scheduled for November 30, 2026. Users are recommended to migrate to Claude Sonnet 5.5.

Claude Sonnet 5.5 (claude-sonnet-5-5) is available on the Claude API, Amazon Bedrock, AWS, Google Cloud, and Microsoft Foundry with a 1M token context window and 128k max output tokens. Code written for Claude Sonnet 5 may break in five ways: disabling up-front thinking now requires thinking: {"type": "between_tools"} instead of "disabled" at effort high or below, forced tool use returns a 400 error, the computer_20251124 computer use tool isn't accepted on Claude API and Google Cloud, and the advisor tool rejects Opus 4.8, 4.7, and Sonnet 5 as advisors.

Billing for refusals before any output resumes for stop_details.category values "bio", "frontier_llm", and "reasoning_extraction", while mid-stream refusals were already billed. Compliance API local session endpoints are out of beta for Claude for Microsoft 365 sessions in Excel, PowerPoint, Word, and Outlook, and Activity Feed no longer returns file names, project document names, or artifact titles — filename and title are now empty or omitted, requiring a Compliance Access Key with read:compliance_user_data scope to look up names by ID.

Compliance API local session endpoints are out of beta for Claude for Microsoft 365 sessions in Excel, PowerPoint, Word, and Outlook, and billing has resumed for refusals arriving before any output when stop_details.category is "bio", "frontier_llm", or "reasoning_extraction". The Compliance API Activity Feed no longer returns file names, project document names, or artifact titles.

Cache diagnostics is out of beta on the Claude API and no longer requires the cache-diagnosis-2026-04-07 beta header; include the diagnostics object on a Messages request to opt in. Responses from POST /v1/messages now always include the diagnostics field, which is null when the request did not include the diagnostics object.

Claude Opus 5.5 (claude-opus-5-5) ships with a 1M token context window by default, 128k max output tokens, and always-on adaptive thinking at $4 / $20 per MTok, down from Opus 5's $5 / $25, across Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. Thinking can no longer be disabled and tool_choice types any and tool now return 400 errors; computer use requires computer_toolset_20260801, and mid-conversation system messages can define tools via the inline-tools-2026-09-15 beta header.

With the inline-tools-2026-09-15 beta header, tools can be defined inside a mid-conversation system message via a tool_addition block carrying the tool's full definition, so a tool can be added, have its schema changed, or be moved to a newer server version without editing tools or invalidating the prompt cache. The same header covers adding and removing tools by reference, and the MCP connector's mcp-client-2026-09-15 header allows an MCP toolset definition with each server's fetched tool list recorded in an mcp_tool_listing block.

Compliance API local session endpoints now return transcripts of Claude in Chrome sessions, identified by a product_surface value of claude_in_chrome. The capability is in beta for Claude Enterprise organizations and uses the existing Compliance Access Key and read:compliance_user_data scope.

The Compliance API local session endpoints now return transcripts of Claude in Chrome sessions (product_surface value claude_in_chrome), in beta for Claude Enterprise organizations. Uses existing Compliance Access Key and read:compliance_user_data scope.

Responses to requests sending the cache-diagnosis-2026-04-07 beta header now always include the diagnostics field, which is null when the request did not include the diagnostics object; previously the field was omitted in that case.