Claude Platform changelog
139 announcements since January 2025 · 15 GA · 18 previews · 2 price changes · 28 deprecations
Free. One email a day, only on days when something you follow has an announcement, or one a week if you switch it in your preferences. You confirm your email with a one-time code; there is no password.
November 20261
October 202615
Fast mode deprecated, removal November 30, 2026; requests with speed fast will then errorDeprecation
Maximum request size raised from 32 MB to 128 MB on first-party Claude API
Tag API requests by end customer so safety actions apply to one customer, not the whole organization
Traces Messages API requests and Managed Agents sessions, with SDK spans in seven languagesPreview
Agents can write dynamic workflows that run many agents in phases and combine resultsPreview
Chat endpoints return unified Claude chats and Docs download as Word files for EnterprisePreview
Launched with 1M token context and 128k max output; Sonnet 5.5 cache reads cut to $0.10 per M tokensGA
adds capabilities.server_tools reporting whether models accept web search and code execution tools
adds capabilities.thinking.types.disabled reporting whether a model accepts thinking set to disabled
adds line field for grouping models; Dreams research preview supports three more models
September 202611
retirement scheduled for November 30, 2026; Sonnet 5.5 recommended for migrationDeprecation
Launches with five breaking changes from Sonnet 5, including thinking and forced tool useDeprecation
Refusals before output are billed again for bio, frontier_llm and reasoning_extraction categoriesPrice
Out of beta, no beta header required; responses now always include diagnostics fieldGA
1M token context, 128k max output, $4 / $20 per MTok, plus inline tools betaGA
Adds ant apply to create and update resources from repository files, with a lockfile
Launched with 1M token context, 128k max output, $10/$50 per MTok, cache reads $0.25 per MTokGA
August 202611
Files and Skills beta headers dropped in SDKs; Console adds personal and service account keys
Session endpoints leave beta; Admin API added to ant CLI and seven SDKs
Moves to httpx2, requires Python 3.10, removes Text Completions API and sampling parametersDeprecation
Computer use, Files API and Skills leave beta; browser use toolset and domain restrictions addedGA
Workbench renamed playground; supports every Messages API parameter and adds feature templates
Returns transcripts of local Cowork and Claude Code sessions for Claude Enterprise, in betaPreview
$2 / $10 per MTok becomes standard price; September 1 increase to $3 / $15 cancelledPrice
Returns Cowork session transcripts for Claude Enterprise organizationsPreview
July 20269
Workbench access ends and experimental prompt tools endpoints are removed on August 17, 2026Deprecation
Available on Claude Fable 5, Mythos 5 and Opus 4.8 with no beta header, correcting earlier notesGA
Manages members, invites, groups and custom roles, in beta for Enterprise organizationsPreview
Adds Claude Fable 5 and Claude Sonnet 5 support; CMEK preservation event docs expanded
API keys and Admin API keys can be given an expiration at creation
agent-memory-2026-07-22 changes memory listing order and replaces managed-agents-2026-04-01Deprecation
June 202612
Launched at $2/$10 per MTok with 1M context window; manual extended thinking removedGA
Fast mode removed; requests run at standard speed and rates without errorDeprecation
Sonnet and Haiku limits match Opus at every tier; tiers consolidated into Start, Build, Scale
Deprecated, removal July 24, 2026; migrate to fast mode for Claude Opus 4.8Deprecation
Management API moved from /v1/organizations/tunnels to /v1/tunnels on the Claude APIDeprecation
SDKs support code_execution_20260120 with REPL state persistence, no beta header required
Retired on the Claude API; requests now return an errorDeprecation
New tool versions disclose 90-second cell limit and add response_inclusion parameter
GET /v1/environments/{id}/work endpoint lists pending work for self-hosted sandbox
Launched with 1M-token context, 128k max output, always-on adaptive thinking and safety classifiersGA
Deprecated, with retirement on the Claude API scheduled for August 5, 2026Deprecation
Adds max_tokens to cap advisor output; refusals with no output are no longer billed
May 202610
Launches with 1M-token context, 128k max output and mid-conversation system messagesDeprecation
Returns richer SEC filing data for financial research and due-diligence workflows
Beta reports cache_miss_reason on Messages requests via diagnostics.previous_message_idPreview
Research preview adds Opus 4.7 with faster output at premium pricing, waitlist accessPreview
Short-lived OIDC tokens from your identity provider replace long-lived static API keysGA
April 202610
1M token context beta retired; requests over 200k tokens now return an errorDeprecation
Lets administrators query rate limits set for organization and workspacesGA
Memory added for managed agents under the managed-agents-2026-04-01 headerPreview
Model claude-3-haiku-20240307 retired; requests return an error, Claude Haiku 4.5 recommendedDeprecation
Launched at $5/$25 per MTok, with task budgets in beta, high-resolution images and xhigh effortGA
Deprecated, with retirement on the Claude API scheduled for June 15, 2026Deprecation
Pairs a faster executor model with a higher-intelligence advisor model for guidance mid-generationPreview
Managed agent harness in beta with sandboxing and tools, plus a Claude API command-line clientPreview
Gated research preview for defensive cybersecurity; Messages API on Amazon Bedrock in us-east-1Preview
March 20264
max_tokens raised to 300k for Opus 4.6 and Sonnet 4.6; 1M context beta retires April 30, 2026Deprecation
Returns max_input_tokens, max_tokens and a capabilities object for each model
New display field omits thinking content from responses for faster streaming; billing unchanged
Out of beta for Opus 4.6 and Sonnet 4.6 at standard pricing; media limit raised to 600GA
February 20264
Automatic caching added; Sonnet 3.7 and Haiku 3.5 retired, Haiku 3 retires April 20, 2026Deprecation
Balanced model with extended thinking, 1M token context (beta); code execution free with web searchGA
Fast mode in research preview for Claude Opus 4.6, up to 2.5x as fast
Claude Opus 4.6 launched, with compaction API in beta and data residency controls