Anthropic · 21 verified changes
Official source: Claude Platform release notes
Organizations can set an IP allowlist in the Claude Console, in beta. Requests to covered services from addresses not on the list are refused, which limits where a leaked API key works. It takes IPv4 and IPv6 addresses or CIDR ranges, up to 1,000 per organization.
Text these models generate carries Anthropic's text watermark on the platforms Anthropic lists. On Claude Sonnet 4.6, text from a request that sets temperature, top_p or top_k to a non-default value might not carry it.
With the message-threads-2026-08-12 beta header the API keeps the conversation server-side, so later requests send only new messages. A thread expires 5 minutes after its last request.
After November 30, 2026, requests to claude-opus-4-8 with speed: fast return an error. Move to Claude Opus 5.5 to keep fast mode.
The Usage and Cost API and the usage and cost report endpoints of the Claude Enterprise Analytics API are now legacy but still supported. Their replacement, the Usage and Cost Analytics API (beta header analytics-usage-cost-2026-08-24), reports token usage and cost by model, workspace, API key or user; Anthropic publishes migration guides.
The maximum request size for the Messages API and the Token Counting API rose from 32 MB to 128 MB on the first-party Claude API.
Claude Models on Google Cloud (Anthropic managed) is the Claude API run by Anthropic and reached through your Google Cloud project: Google Cloud IAM controls access and Google Cloud Marketplace handles billing. Fast mode for Claude Opus 5.5 is available there too.
Anthropic released API version 2026-10-21 and new SDK and ant CLI versions that send it by default (Python 2.0.0, TypeScript 1.0.0, C# 13.0.0, Go 2.0.0, Java 3.0.0, PHP 1.0.0, Ruby 2.0.0, CLI 2.0.0). The SDKs also carry breaking changes unrelated to the API version and move from MIT to Apache 2.0.
If you run a platform or gateway, or resell API access, you can tell Anthropic which end customer each request is for, so a safety enforcement action can apply to that customer instead of your whole organization. Since 21 Oct, profiles also show enforcement_status and enforcement_actions.
Requests to Claude Opus 5.5 no longer require access approval or the fast-mode-2026-02-01 beta header; a default fast-mode rate limit applies in Start, Build and Scale tiers. Fast mode for Opus 5 / 4.8 remains in research preview.
You can trace Messages API requests and Claude Managed Agents sessions in your observability backend; the Claude SDK emits spans in Python, TypeScript, C#, Go, Java, PHP and Ruby after adding an OpenTelemetry destination in the Console.
claude-haiku-5-5 has a 1M-token context, 128k max output and adaptive thinking on by default. On it, budget_tokens thinking, temperature/top_p/top_k, assistant prefill and computer_20250124 can return 400 errors, as can resending thinking blocks after changing system, tools or earlier turns. Check Anthropic's migration guide.
To reduce data-exfiltration risk, web_fetch in Claude Managed Agents only fetches URLs that appeared earlier in the session (a user message, a web_search result or a fetched page). Others, such as URLs in Claude's output, the system prompt or tool output, return url_not_in_prior_context; send them in a user.message instead.
With limited networking, web_search and web_fetch now follow allowed_hosts (fetches elsewhere return url_not_allowed), and creating or updating a session fails with a 400 error when a web tool's allowed_domains has an entry outside allowed_hosts. Add the host to allowed_hosts or drop the entry.
Anthropic lowered the prompt-cache read price for Claude Sonnet 5.5 to 0.05x the base input price (from 0.1x). Cache writes and all other prices are unchanged.
Anthropic's Claude Max and Team subscription plans now include monthly Claude API credits; the documentation page on API credits for Max and Team plans explains how to claim them.
Anthropic announced the deprecation of claude-sonnet-4-5-20250929, with retirement on the Claude API scheduled for 2026-11-30, and recommends migrating to Claude Sonnet 5.5.
Launch of claude-sonnet-5-5. Breaking notes include: forced tool use (tool_choice any/tool) returns 400, and to turn off up-front thinking you send thinking {type: 'between_tools'} instead of 'disabled' at high effort or below. See Anthropic's migration guide before swapping model strings.
The filename and title fields on file, project document and artifact activities are now always empty or omitted, including for past activities. To look up a name or title, use a Compliance Access Key with the read:compliance_user_data scope.
Refusals before any output are billed again when stop_details.category is bio, frontier_llm or reasoning_extraction, at the rates of the model that ran. Mid-stream refusals were already billed; early refusals in other categories stay free, and fallback credit is unchanged. Applies on all platforms.
claude-opus-5-5: 1M-token context by default, 128k max output, always-on adaptive thinking, priced $4 input / $20 output per MTok (Opus 5 is $5/$25). thinking {type:'disabled'} or {'enabled'} returns a 400 error; omit thinking and control depth with effort.