◆ AI Model & Data Infrastructure

Anthropic

August 27, 2026 commercial

Anthropic is an AI safety company providing Claude, a large language model AI assistant for text generation, analysis, and conversation.

source ↗ rss ↗

curl https://api.anthropic.com/v1/messages \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'content-type: application/json' \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 1024,
    "system": [
      {
        "type": "text",
        "text": "<large system prompt>",
        "cache_control": {"type": "ephemeral", "ttl": "1h"}
      }
    ],
    "messages": [{"role": "user", "content": "Summarize the above."}]
  }'
curl https://api.anthropic.com/v1/messages \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'content-type: application/json' \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 0,
    "cache_control": {"type": "ephemeral", "ttl": "1h"},
    "messages": [{"role": "user", "content": "<large shared context>"}]
  }'
curl https://api.anthropic.com/v1/messages \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'content-type: application/json' \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 512,
    "output_config": {
      "format": {
        "type": "json_schema",
        "schema": {
          "type": "object",
          "properties": {
            "severity": {"type": "string"},
            "cve": {"type": "string"},
            "summary": {"type": "string"}
          },
          "required": ["severity", "cve", "summary"]
        }
      }
    },
    "messages": [{"role": "user", "content": "Analyze this vulnerability report: <report>"}]
  }'
import httpx2
httpx2.alias_httpx()

import anthropic
client = anthropic.Anthropic()
import anthropic
import asyncio

async def main():
    client = anthropic.AsyncAnthropic()
    raw = await client.messages.with_raw_response.create(
        model='claude-opus-4-5',
        max_tokens=256,
        messages=[{'role': 'user', 'content': 'Hello'}]
    )
    message = await raw.parse()
    print(message.content)

asyncio.run(main())
curl https://api.anthropic.com/v1/files \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -F '[email protected];type=application/pdf' \
  -F 'expires_in_seconds=86400'
curl -X POST https://api.anthropic.com/v1/agents \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": {"name": "claude-opus-5"},
    "agent_toolset_20260401": {
      "configs": [
        {
          "name": "web_search",
          "type": "web_search",
          "allowed_domains": ["docs.anthropic.com", "en.wikipedia.org"],
          "user_location": {"country": "US"}
        },
        {
          "name": "web_fetch",
          "type": "web_fetch",
          "allowed_domains": ["docs.anthropic.com", "en.wikipedia.org"],
          "max_content_tokens": 8000
        }
      ]
    }
  }'
curl https://api.anthropic.com/v1/files \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -F '[email protected];type=application/pdf' \
  -F 'expires_in_seconds=86400'
curl -X POST https://api.anthropic.com/v1/agents \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": {"name": "claude-opus-5"},
    "agent_toolset_20260401": {
      "configs": [
        {
          "name": "web_search",
          "type": "web_search",
          "allowed_domains": ["docs.anthropic.com", "en.wikipedia.org"],
          "user_location": {"country": "US"}
        },
        {
          "name": "web_fetch",
          "type": "web_fetch",
          "allowed_domains": ["docs.anthropic.com", "en.wikipedia.org"],
          "max_content_tokens": 8000
        }
      ]
    }
  }'
import httpx2
httpx2.alias_httpx()

import anthropic
client = anthropic.Anthropic()
import anthropic
import asyncio

async def main():
    client = anthropic.AsyncAnthropic()
    raw = await client.messages.with_raw_response.create(
        model="claude-opus-4-5",
        max_tokens=256,
        messages=[{"role": "user", "content": "Hello"}]
    )
    message = await raw.parse()
    print(message.content)

asyncio.run(main())
{
  "agent_toolset": "agent_toolset_20260401",
  "configs": [
    {
      "name": "web_search",
      "type": "web_search",
      "allowed_domains": ["example.com", "docs.example.com"],
      "user_location": "US"
    },
    {
      "name": "web_fetch",
      "type": "web_fetch",
      "allowed_domains": ["example.com"],
      "max_content_tokens": 8000
    }
  ]
}
# Upload a file with a 7-day expiry
curl https://api.anthropic.com/v1/files \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -F '[email protected];type=application/pdf' \
  -F 'expires_in_seconds=604800'

# List specific files by ID with pagination
curl 'https://api.anthropic.com/v1/files?ids[]=file_abc123&ids[]=file_def456&page=1' \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01'
curl https://api.anthropic.com/v1/compliance/apps/sessions/local/session_abc123/messages \
  -H 'Authorization: Bearer <compliance_access_key>' \
  -H 'anthropic-version: 2023-06-01'
curl -I https://api.anthropic.com/v1/messages \
  -H 'x-api-key: <api_key>' \
  -H 'anthropic-version: 2023-06-01' \
  -d '{"model":"claude-opus-4-5","max_tokens":1,"messages":[{"role":"user","content":"hi"}]}' \
  | grep anthropic-workspace-id
{
  "multiagent": [
    { "type": "advisor", "model": "claude-opus-5" }
  ]
}
{
  "model": {
    "name": "claude-opus-5",
    "inference_geo": "eu"
  }
}
curl -X GET 'https://platform.claude.com/v1/compliance/apps/sessions/remote/<session_id>/messages' \
  -H 'Authorization: Bearer <compliance_access_key>'
curl -X GET 'https://platform.claude.com/v1/compliance/apps/sessions/remote' \
  -H 'Authorization: Bearer <compliance_access_key>'

Summary

Anthropic is an AI safety company providing Claude, a large language model AI assistant for text generation, analysis, and conversation.

Release history

docs update Aug 28, 2026 · issue 009

Anthropic beta Sessions API adds full CRUD plus multiagent, memory-store, and budget-limit capabilities for managed agent sessions.
- ›Adds POST /v1/sessions to create a new managed agent session.
- ›Adds GET /v1/sessions to list all sessions in the caller's workspace.
- ›Adds GET /v1/sessions/{session_id} to retrieve a specific session.
- ›Adds POST /v1/sessions/{session_id} to update a session mid-run; only tools and mcp_servers fields on the agent are updatable, with full-replacement semantics.
- ›Adds DELETE /v1/sessions/{session_id} to permanently delete a session, returning a session_deleted confirmation object.
+15 moreshow less
- ›Adds POST /v1/sessions/{session_id}/archive to archive a session.
- ›Introduces BetaManagedAgentsBudgetLimit with max_list_cost (amount + ISO-4217 currency, USD only) as a hard spend ceiling — the session stops issuing new model requests once tracked list cost reaches the limit.
- ›Introduces BetaManagedAgentsMemoryStoreResourceParam to attach a memory store (memstore_... ID) to a session with read_write or read_only access and per-attachment instructions (max 4096 chars) rendered into the system prompt.
- ›Introduces BetaManagedAgentsFileResourceParams to mount a Files API upload into the session container at an optional mount_path (defaults to /mnt/session/uploads/<file_id>).
- ›Introduces BetaManagedAgentsGitHubRepositoryResourceParams to mount a GitHub repository into the session container.
- ›Supports multiagent coordinator topology via BetaManagedAgentsMultiagentParams, where the primary thread orchestrates subagents drawn from a named roster, with entries typed as agent ID strings, versioned BetaManagedAgentsAgentParams references, self, or BetaManagedAgentsAdvisorParams (occupies roster name anthropic.advisor).
- ›Adds streaming delta events (BetaManagedAgentsStartEvent / BetaManagedAgentsDeltaEvent / BetaManagedAgentsDeltaContent) for agent.message and agent.thinking previews on stream connections that opt in via event_deltas.
- ›Adds BetaManagedAgentsSessionUsageEvent for periodic cumulative token-usage and tracked list-cost snapshots, and BetaManagedAgentsSessionStats exposing active_seconds and duration_seconds.
- ›Adds BetaManagedAgentsCacheCreationUsage breaking down prompt-cache creation by lifetime: ephemeral_1h_input_tokens and ephemeral_5m_input_tokens.
- ›Adds BetaManagedAgentsServerToolUsage tracking cumulative web_fetch_requests and web_search_requests across a session.
- ›Adds BetaManagedAgentsOutcomeEvaluationResource to track grader-scored outcomes defined via define_outcome events, with states pending, running, evaluating, satisfied, needs_revision, max_iterations_reached, failed, and interrupted.
- ›Adds BetaManagedAgentsSystemMessageEvent (system.message type) allowing mid-conversation system-role content to be appended to a session as a role: 'system' turn.
- ›Adds BetaManagedAgentsUserToolResultEvent for client-side tool result delivery back to the session.
- ›Adds BetaManagedAgentsSessionUpdatedEvent emitted when UpdateSession changes at least one field, carrying only the changed fields; new configuration applies from the next turn.
- ›Supports BetaManagedAgentsBranchCheckout and BetaManagedAgentsCommitCheckout for pinning repository state to a branch name or full commit SHA.
docs update Aug 28, 2026 · issue 009

Anthropic Messages API gains new models, container skills, prompt-cache TTL control, user-profile attribution, and structured output configuration.
└──▷ TRY IT
Cache a large system prompt for 1 hour to reduce costs on repeated requests with the same context.
```
$ curl https://api.anthropic.com/v1/messages \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'content-type: application/json' \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 1024,
    "system": [
      {
        "type": "text",
        "text": "<large system prompt>",
        "cache_control": {"type": "ephemeral", "ttl": "1h"}
      }
    ],
    "messages": [{"role": "user", "content": "Summarize the above."}]
  }'
```
Pre-warm the prompt cache without generating output — useful for seeding expensive context before a burst of requests.
```
$ curl https://api.anthropic.com/v1/messages \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'content-type: application/json' \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 0,
    "cache_control": {"type": "ephemeral", "ttl": "1h"},
    "messages": [{"role": "user", "content": "<large shared context>"}]
  }'
```
Request a structured JSON response using output_config with a schema — useful for parsing model output programmatically.
```
$ curl https://api.anthropic.com/v1/messages \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'content-type: application/json' \
  -d '{
    "model": "claude-sonnet-5",
    "max_tokens": 512,
    "output_config": {
      "format": {
        "type": "json_schema",
        "schema": {
          "type": "object",
          "properties": {
            "severity": {"type": "string"},
            "cve": {"type": "string"},
            "summary": {"type": "string"}
          },
          "required": ["severity", "cve", "summary"]
        }
      }
    },
    "messages": [{"role": "user", "content": "Analyze this vulnerability report: <report>"}]
  }'
```
- ›Adds anthropic-user-profile-id request header (requires user-profiles beta header) to attribute API requests to a specific user profile when acting on behalf of a party other than your organization.
- ›Adds ttl field to CacheControlEphemeral objects ('5m' or '1h') to control prompt-cache time-to-live per content block; defaults to '5m'.
- ›Adds top-level cache_control parameter to automatically apply a cache breakpoint to the last cacheable block in the request, with the same ttl options.
- ›Adds max_tokens: 0 support to populate the prompt cache without generating a response.
- ›Adds container parameter (ContainerParams with id and skills) to reuse a container across requests and load up to 20 named skills (type 'anthropic' or 'custom', optional version).
+8 moreshow less
- ›Adds output_config parameter with effort ('low', 'medium', 'high', 'xhigh', 'max') and format (JSONOutputFormat with schema and type: 'json_schema') for structured output control.
- ›Adds inference_geo parameter to pin inference processing to a specific geographic region per request; falls back to the workspace's default_inference_geo when omitted.
- ›Adds service_tier parameter ('auto' or 'standard_only') to choose between priority and standard capacity per request.
- ›Adds ContainerUploadBlockParam content block type (type: 'container_upload', file_id) to upload files into a container's input directory as part of a message.
- ›Adds ToolSearchToolResultBlockParam, BashCodeExecutionToolResultBlockParam, and TextEditorCodeExecutionToolResultBlockParam content block types for richer code-execution tool results.
- ›Adds new model identifiers: claude-sonnet-5 (coding/agents), claude-fable-5 (hard knowledge/coding), claude-mythos-5 (cybersecurity and biology research), claude-opus-5 (long-running agents), and claude-mythos-preview (strongest in coding and cybersecurity).
- ›Adds 'system' as an accepted value for the role field in MessageParam.
- ›Extends messages limit to 100,000 messages per single request.
August 27, 2026 Aug 27, 2026 · issue 009

Claude API adds personal and service account API keys with workspace scoping; Files and Skills APIs graduate out of beta.
- ›Adds personal keys and service account keys in the Claude Console, scoped to a specific workspace or admin endpoints across any workspace the linked account can access — enabling per-account usage tracking and legitimacy enforcement by org admins.
- ›client.beta.files and client.beta.skills no longer send the files-api-2025-04-14 and skills-2025-10-02 beta headers across Python SDK 1.2.0, TypeScript SDK 0.122.0, Go SDK 1.68.0, Java SDK 2.59.0, Ruby SDK 1.67.0, and C# SDK 12.44.0 — the Files and Skills APIs are now GA.
- ›client.beta.skills.delete() now deletes a Skill together with all of its versions (previously deleted only the referenced version under the beta header).
└──▷ BREAKING ON UPGRADE
- !The beta Messages type BetaSkill is renamed to BetaContainerSkill in the SDKs listed (Python 1.2.0, TypeScript 0.122.0, Go 1.68.0, Java 2.59.0, Ruby 1.67.0, C# 12.44.0); code referencing BetaSkill by name will break.
- !client.beta.skills.delete() now deletes a Skill and ALL of its versions — callers expecting single-version deletion under the old beta behavior will see additional versions removed.
- !client.beta.files and client.beta.skills now return the GA response shapes (same as client.files and client.skills); code that relied on the beta-specific response shapes will break unless it still sends the beta headers explicitly.
August 26, 2026 Aug 26, 2026 · issue 009

Compliance API adds Claude Science and Office 365 surfaces; Admin API lands in CLI and seven SDKs
- ›Compliance API session endpoints for Cowork and Claude Code sessions are now generally available (out of beta); see Retrieve session transcripts.
- ›Compliance API local session endpoints now return transcripts for Claude Science sessions via product_surface value claude_science, in beta for Claude Enterprise organizations.
- ›Compliance API local session endpoints now return transcripts for Claude for Microsoft 365 sessions in Excel, PowerPoint, Word, and Outlook via product_surface values beginning with office_agents, in beta for Claude Enterprise organizations, using the existing Compliance Access Key and read:compliance_user_data scope.
- ›Admin API is now available in the ant CLI and the Python, TypeScript, C#, Go, Java, PHP, and Ruby SDKs under client.beta.organization, covering organization info, members, invites, workspaces and workspace members, API keys, rate limits, service accounts, workload identity federation issuers and rules, and customer-managed encryption keys.
- ›The ant CLI and SDKs read an Admin API key from ANTHROPIC_API_KEY or an org:admin OAuth token from ANTHROPIC_AUTH_TOKEN.
snapshot-20260822 seen Aug 22, 2026 · issue 004

Python SDK v1.0 moves HTTP layer to httpx2 and drops long-deprecated surfaces including the legacy Text Completions API.
└──▷ USE IT
Preserve existing tracing or mocking libraries that patch httpx after migrating to the httpx2-backed SDK.
python
```
import httpx2
httpx2.alias_httpx()

import anthropic
client = anthropic.Anthropic()
```
Correctly await a raw response parse on the async client under SDK v1.0.
python
```
import anthropic
import asyncio

async def main():
    client = anthropic.AsyncAnthropic()
    raw = await client.messages.with_raw_response.create(
        model='claude-opus-4-5',
        max_tokens=256,
        messages=[{'role': 'user', 'content': 'Hello'}]
    )
    message = await raw.parse()
    print(message.content)

asyncio.run(main())
```
- ›Requires Python 3.10 or later.
- ›On the async client, .with_raw_response results now use await response.parse() instead of the previous synchronous parse call.
└──▷ BREAKING ON UPGRADE
- !The legacy Text Completions API is removed.
- !The temperature, top_p, and top_k parameters on Messages methods are removed.
- !The tool runner's client-side compaction_control is removed.
- !On the async client, .with_raw_response results now require await response.parse() — synchronous .parse() no longer works.
- !AnthropicBedrock now raises an error when no AWS region is configured instead of defaulting to us-east-1.
- !Custom http_client, Timeout, and transport objects must now be built from httpx2, not httpx; the DefaultHttpxClient helpers are unchanged.
snapshot-20260821 seen Aug 21, 2026 · issue 003

Claude API adds GA computer use toolset, new browser use toolset, and GA Admin API user-management endpoints for Enterprise orgs.
- ›Promotes the computer use tool to general availability as computer_toolset_20260801, requiring no beta header; adds batch actions (multiple actions per turn), zoom enabled by default, and per-member configuration via configs.
- ›Introduces browser_toolset_20260801, a new client toolset for driving an application-hosted browser viewport; reads the page accessibility tree, elements, forms, and tabs, and adds element references, form input, tab management, download reporting, and opt-in file upload on top of screenshot-and-click control.
└──▷ BREAKING ON UPGRADE
- !Upgrading an existing computer use integration to computer_toolset_20260801 changes the request shape and tool handling; migration from computer_20251124 is required (see 'Migrate from computer_20251124').
snapshot-20260821 seen Aug 21, 2026 · issue 001

Claude API adds GA computer use toolset, new browser use toolset, and GA Admin API user-management endpoints for Enterprise orgs.
- ›Promotes the computer use tool to general availability as computer_toolset_20260801, requiring no beta header; adds batch actions (multiple actions per turn), zoom enabled by default, and per-member configuration via configs.
- ›Introduces browser_toolset_20260801, a new client toolset for driving an application-hosted browser viewport; reads the page accessibility tree, elements, forms, and tabs, and adds element references, form input, tab management, download reporting, and opt-in file upload on top of screenshot-and-click control.
└──▷ BREAKING ON UPGRADE
- !Upgrading an existing computer use integration to computer_toolset_20260801 changes the request shape and tool handling; migration from computer_20251124 is required (see 'Migrate from computer_20251124').
snapshot-20260820 seen Aug 20, 2026 · issue 002

Claude API reaches GA for Files, Skills, and Enterprise user management; Managed Agents gains budgets, advisors, memory stores, domain restrictions, and more.
└──▷ TRY IT
Upload a file with a TTL and then reference it in a Messages API request without the old beta header.
```
$ curl https://api.anthropic.com/v1/files \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -F '[email protected];type=application/pdf' \
  -F 'expires_in_seconds=86400'
```
Restrict a web-search-enabled agent to only fetch from approved domains, preventing lateral browsing to untrusted sites.
```
$ curl -X POST https://api.anthropic.com/v1/agents \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": {"name": "claude-opus-5"},
    "agent_toolset_20260401": {
      "configs": [
        {
          "name": "web_search",
          "type": "web_search",
          "allowed_domains": ["docs.anthropic.com", "en.wikipedia.org"],
          "user_location": {"country": "US"}
        },
        {
          "name": "web_fetch",
          "type": "web_fetch",
          "allowed_domains": ["docs.anthropic.com", "en.wikipedia.org"],
          "max_content_tokens": 8000
        }
      ]
    }
  }'
```
- ›Admin API user-management endpoints (members, invites, groups, custom roles) for Claude Enterprise organizations are now GA; the anthropic-beta: ce-user-management-2026-07-13 beta header is no longer required on group and custom-role requests.
- ›Files API is now GA on /v1/files; the files-api-2025-04-14 beta header is no longer required. GA response format adds expires_in_seconds (upload) and expires_at (file objects), page and next_page pagination, and an ids[] filter on list requests. Storage limit is 1 TB per organization; rate limit is 500 requests per minute.
- ›Agent Skills and the Skills API (/v1/skills) are now GA; the skills-2025-10-02 beta header is no longer required, including for Messages API requests that load Skills through the container parameter.
- ›Adds allowed_domains and blocked_domains controls on web_search and web_fetch tool entries in the agent_toolset_20260401 configs array for Claude Managed Agents; web_fetch also accepts max_content_tokens and web_search accepts user_location.
- ›Claude Managed Agents sessions running in a self-hosted sandbox can now attach memory stores; Python, TypeScript, and Go SDK workers download each store to its mount_path and sync changes back.
+18 moreshow less
- ›Redesigned session viewer in the Claude Console adds a timeline minimap, transcript grouped by model request, and an Inspector panel covering session details and cost, raw events, per-tool statistics, mounted resources, and per-thread activity.
- ›Workbench is now Playground at platform.claude.com/playground; supports every Messages API parameter, includes feature templates (code execution, web search), and shows the full SDK request and API response for each run.
- ›Compliance API (beta for Claude Enterprise) adds GET /v1/compliance/apps/sessions/local to list local Cowork and Claude Code sessions, GET /v1/compliance/apps/sessions/local/{session_id} for session metadata, and GET /v1/compliance/apps/sessions/local/{session_id}/messages for transcripts, using the existing Compliance Access Key with read:compliance_user_data scope.
- ›Adds anthropic-workspace-id response header to the Claude API, carrying the wrkspc_-prefixed ID of the workspace the request resolved to.
- ›Adds session budget support for Claude Managed Agents: set a hard spend cap per session; sessions that reach the cap pause with the budget_reached stop reason; deployments accept the same budget field and apply it to every session they start.
- ›Adds advisor support for Claude Managed Agents sessions: configure a {"type": "advisor"} entry in the agent's multiagent roster to give the primary thread a model to consult mid-turn for strategic guidance.
- ›Adds inference_geo inside the model object when creating a Claude Managed Agents agent, or as a per-session override, to control where model inference runs.
- ›Claude Managed Agents sessions can now load skills from a GitHub repository; skills placed in the repository's root .claude/skills directory are discovered automatically at session start.
- ›Inference hooks are now in beta for Claude Enterprise organizations: point Claude at an AI security server to hold governed prompts from claude.ai, Cowork, and Claude Code for allow/deny verdicts before inference; denials are recorded in the compliance Activity Feed.
- ›Compliance API (beta for Claude Enterprise) adds GET /v1/compliance/apps/sessions/remote to list cloud Cowork sessions and GET /v1/compliance/apps/sessions/remote/{session_id}/messages for transcripts.
- ›Launches Claude Opus 5 (claude-opus-5) with a 1 M token context window, 128k max output tokens, thinking on by default, and full effort ladder (low, medium, high, xhigh, max), at $5 / $25 per MTok.
- ›Mid-conversation tool changes are now in beta on Claude Fable 5, Claude Mythos 5, Claude Opus 4.8, and Claude Opus 5: add or remove tools between turns while preserving the prompt cache using the mid-conversation-tool-changes-2026-07-01 beta header.
- ›The fallbacks parameter now supports a "default" mode applying Anthropic's recommended fallback models by refusal category; requires the server-side-fallback-2026-07-01 beta header.
- ›Adds effort field inside the model object when creating a Claude Managed Agents agent.
- ›Claude Managed Agents webhooks now cover four environment.* event types and three memory_store.* event types for environment and memory store lifecycle changes.
- ›Adds initial_events parameter on POST /v1/sessions (up to 50 user.message and user.define_outcome events) to seed a Claude Managed Agents session and start the agent loop in the same call.
- ›The version field is now optional when updating a Claude Managed Agents agent via the update endpoint; omit it to apply unconditionally, or include it for optimistic concurrency (mismatch returns 409).
- ›Claude Managed Agents session thread event stream (GET /v1/sessions/{session_id}/threads/{thread_id}/stream) now accepts the event_deltas[] query parameter to preview subagent text as the model generates it.
└──▷ BREAKING ON UPGRADE
- !On Claude Opus 5, thinking: {"type": "disabled"} combined with effort of xhigh or max returns a 400 error; this was allowed on Claude Opus 4.8.
- !Requests to claude-opus-4-7 with speed: "fast" now return an error; unlike Claude Opus 4.6, they do not fall back to standard speed.
- !Claude Opus 4.1 (claude-opus-4-1-20250805) has been retired; all API requests to this model now return an error.
- !The experimental prompt tools APIs (/v1/experimental/generate_prompt, /v1/experimental/improve_prompt, /v1/experimental/templatize_prompt) are being retired on August 17, 2026.
- !The legacy Workbench (platform.claude.com/workbench) is being sunset on August 17, 2026; saved prompts, variables, and evals are not supported in the updated Playground.
snapshot-20260820 seen Aug 20, 2026 · issue 001

Claude API reaches GA for Files, Skills, and Enterprise user management; Managed Agents gains budgets, advisors, memory stores, domain restrictions, and more.
└──▷ TRY IT
Upload a file with a TTL and then reference it in a Messages API request without the old beta header.
```
$ curl https://api.anthropic.com/v1/files \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -F '[email protected];type=application/pdf' \
  -F 'expires_in_seconds=86400'
```
Restrict a web-search-enabled agent to only fetch from approved domains, preventing lateral browsing to untrusted sites.
```
$ curl -X POST https://api.anthropic.com/v1/agents \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -H 'Content-Type: application/json' \
  -d '{
    "model": {"name": "claude-opus-5"},
    "agent_toolset_20260401": {
      "configs": [
        {
          "name": "web_search",
          "type": "web_search",
          "allowed_domains": ["docs.anthropic.com", "en.wikipedia.org"],
          "user_location": {"country": "US"}
        },
        {
          "name": "web_fetch",
          "type": "web_fetch",
          "allowed_domains": ["docs.anthropic.com", "en.wikipedia.org"],
          "max_content_tokens": 8000
        }
      ]
    }
  }'
```
- ›Admin API user-management endpoints (members, invites, groups, custom roles) for Claude Enterprise organizations are now GA; the anthropic-beta: ce-user-management-2026-07-13 beta header is no longer required on group and custom-role requests.
- ›Files API is now GA on /v1/files; the files-api-2025-04-14 beta header is no longer required. GA response format adds expires_in_seconds (upload) and expires_at (file objects), page and next_page pagination, and an ids[] filter on list requests. Storage limit is 1 TB per organization; rate limit is 500 requests per minute.
- ›Agent Skills and the Skills API (/v1/skills) are now GA; the skills-2025-10-02 beta header is no longer required, including for Messages API requests that load Skills through the container parameter.
- ›Adds allowed_domains and blocked_domains controls on web_search and web_fetch tool entries in the agent_toolset_20260401 configs array for Claude Managed Agents; web_fetch also accepts max_content_tokens and web_search accepts user_location.
- ›Claude Managed Agents sessions running in a self-hosted sandbox can now attach memory stores; Python, TypeScript, and Go SDK workers download each store to its mount_path and sync changes back.
+18 moreshow less
- ›Redesigned session viewer in the Claude Console adds a timeline minimap, transcript grouped by model request, and an Inspector panel covering session details and cost, raw events, per-tool statistics, mounted resources, and per-thread activity.
- ›Workbench is now Playground at platform.claude.com/playground; supports every Messages API parameter, includes feature templates (code execution, web search), and shows the full SDK request and API response for each run.
- ›Compliance API (beta for Claude Enterprise) adds GET /v1/compliance/apps/sessions/local to list local Cowork and Claude Code sessions, GET /v1/compliance/apps/sessions/local/{session_id} for session metadata, and GET /v1/compliance/apps/sessions/local/{session_id}/messages for transcripts, using the existing Compliance Access Key with read:compliance_user_data scope.
- ›Adds anthropic-workspace-id response header to the Claude API, carrying the wrkspc_-prefixed ID of the workspace the request resolved to.
- ›Adds session budget support for Claude Managed Agents: set a hard spend cap per session; sessions that reach the cap pause with the budget_reached stop reason; deployments accept the same budget field and apply it to every session they start.
- ›Adds advisor support for Claude Managed Agents sessions: configure a {"type": "advisor"} entry in the agent's multiagent roster to give the primary thread a model to consult mid-turn for strategic guidance.
- ›Adds inference_geo inside the model object when creating a Claude Managed Agents agent, or as a per-session override, to control where model inference runs.
- ›Claude Managed Agents sessions can now load skills from a GitHub repository; skills placed in the repository's root .claude/skills directory are discovered automatically at session start.
- ›Inference hooks are now in beta for Claude Enterprise organizations: point Claude at an AI security server to hold governed prompts from claude.ai, Cowork, and Claude Code for allow/deny verdicts before inference; denials are recorded in the compliance Activity Feed.
- ›Compliance API (beta for Claude Enterprise) adds GET /v1/compliance/apps/sessions/remote to list cloud Cowork sessions and GET /v1/compliance/apps/sessions/remote/{session_id}/messages for transcripts.
- ›Launches Claude Opus 5 (claude-opus-5) with a 1 M token context window, 128k max output tokens, thinking on by default, and full effort ladder (low, medium, high, xhigh, max), at $5 / $25 per MTok.
- ›Mid-conversation tool changes are now in beta on Claude Fable 5, Claude Mythos 5, Claude Opus 4.8, and Claude Opus 5: add or remove tools between turns while preserving the prompt cache using the mid-conversation-tool-changes-2026-07-01 beta header.
- ›The fallbacks parameter now supports a "default" mode applying Anthropic's recommended fallback models by refusal category; requires the server-side-fallback-2026-07-01 beta header.
- ›Adds effort field inside the model object when creating a Claude Managed Agents agent.
- ›Claude Managed Agents webhooks now cover four environment.* event types and three memory_store.* event types for environment and memory store lifecycle changes.
- ›Adds initial_events parameter on POST /v1/sessions (up to 50 user.message and user.define_outcome events) to seed a Claude Managed Agents session and start the agent loop in the same call.
- ›The version field is now optional when updating a Claude Managed Agents agent via the update endpoint; omit it to apply unconditionally, or include it for optimistic concurrency (mismatch returns 409).
- ›Claude Managed Agents session thread event stream (GET /v1/sessions/{session_id}/threads/{thread_id}/stream) now accepts the event_deltas[] query parameter to preview subagent text as the model generates it.
└──▷ BREAKING ON UPGRADE
- !On Claude Opus 5, thinking: {"type": "disabled"} combined with effort of xhigh or max returns a 400 error; this was allowed on Claude Opus 4.8.
- !Requests to claude-opus-4-7 with speed: "fast" now return an error; unlike Claude Opus 4.6, they do not fall back to standard speed.
- !Claude Opus 4.1 (claude-opus-4-1-20250805) has been retired; all API requests to this model now return an error.
- !The experimental prompt tools APIs (/v1/experimental/generate_prompt, /v1/experimental/improve_prompt, /v1/experimental/templatize_prompt) are being retired on August 17, 2026.
- !The legacy Workbench (platform.claude.com/workbench) is being sunset on August 17, 2026; saved prompts, variables, and evals are not supported in the updated Playground.
August 20, 2026 Aug 20, 2026 · issue 005

Python SDK v1.0 ships with httpx2, drops legacy APIs, and tightens async and AWS region handling.
└──▷ USE IT
Preserve httpx-based mocking or tracing libraries after upgrading to the httpx2 HTTP layer.
python
```
import httpx2
httpx2.alias_httpx()

import anthropic
client = anthropic.Anthropic()
```
Await parsed responses correctly on the async client after the v1.0 .with_raw_response change.
python
```
import anthropic
import asyncio

async def main():
    client = anthropic.AsyncAnthropic()
    raw = await client.messages.with_raw_response.create(
        model="claude-opus-4-5",
        max_tokens=256,
        messages=[{"role": "user", "content": "Hello"}]
    )
    message = await raw.parse()
    print(message.content)

asyncio.run(main())
```
- ›Migrates the SDK's HTTP layer from httpx to httpx2; build custom http_client, Timeout, and transport objects from httpx2 — the DefaultHttpxClient helpers are unchanged.
- ›Adds httpx2.alias_httpx() startup call to preserve compatibility with tracing or mocking libraries that patch httpx.
- ›Requires Python 3.10 or later.
- ›Removes the legacy Text Completions API.
- ›Removes the temperature, top_p, and top_k parameters from Messages methods.
+3 moreshow less
- ›Removes the tool runner's client-side compaction_control.
- ›Changes async client .with_raw_response so results now require await response.parse() instead of a synchronous call.
- ›AnthropicBedrock now raises an error when no AWS region is configured, rather than silently defaulting to us-east-1.
└──▷ BREAKING ON UPGRADE
- !The HTTP layer moves from httpx to httpx2; custom http_client, Timeout, and transport objects must now be constructed from httpx2, not httpx.
- !Tracing or mocking libraries that patch httpx will no longer intercept SDK calls unless httpx2.alias_httpx() is called at startup.
- !Python 3.9 and earlier are no longer supported; Python 3.10 or later is required.
- !The legacy Text Completions API is removed.
- !The temperature, top_p, and top_k parameters on Messages methods are removed.
- !The tool runner's compaction_control option is removed.
- !On the async client, .with_raw_response results now require await response.parse() — code that called .parse() synchronously will break.
- !AnthropicBedrock now raises an error when no AWS region is configured, breaking setups that relied on the implicit us-east-1 default.
August 19, 2026 Aug 19, 2026 · issue 005

Computer use and browser use toolsets GA, Files/Skills/Admin APIs out of beta, domain restrictions and memory stores for Managed Agents.
└──▷ USE IT
Restrict a Managed Agents web_search tool to approved domains to prevent agents from reaching untrusted sites.
json
```
{
  "agent_toolset": "agent_toolset_20260401",
  "configs": [
    {
      "name": "web_search",
      "type": "web_search",
      "allowed_domains": ["example.com", "docs.example.com"],
      "user_location": "US"
    },
    {
      "name": "web_fetch",
      "type": "web_fetch",
      "allowed_domains": ["example.com"],
      "max_content_tokens": 8000
    }
  ]
}
```
Upload a file with an expiry and list files filtered by ID using the now-GA Files API without a beta header.
```
$ # Upload a file with a 7-day expiry
curl https://api.anthropic.com/v1/files \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01' \
  -F '[email protected];type=application/pdf' \
  -F 'expires_in_seconds=604800'

# List specific files by ID with pagination
curl 'https://api.anthropic.com/v1/files?ids[]=file_abc123&ids[]=file_def456&page=1' \
  -H 'x-api-key: $ANTHROPIC_API_KEY' \
  -H 'anthropic-version: 2023-06-01'
```
Inspect cost, per-tool statistics, and per-thread activity for a Managed Agents session in the Claude Console.
📍1. Open the Claude Console and navigate to the Managed Agents section. 2. Select a completed or active session to open the redesigned session viewer. 3. Use the timeline minimap to jump to a point of interest in the session. 4. Click the Inspector panel to view session cost, raw events, per-tool statistics, mounted resources, and per-thread activity.
- ›Promotes computer_toolset_20260801 to GA on the Claude API — no beta header required, adds batch actions (multiple actions per turn), enables zoom by default, and supports per-member configuration via configs; earlier beta versions remain available.
- ›Promotes the Files API (/v1/files endpoints) to GA — beta header files-api-2025-04-14 no longer required; new response format adds expires_at on file objects, expires_in_seconds on upload, and page/next_page pagination plus an ids[] filter on list requests.
- ›Promotes Agent Skills and the Skills API (/v1/skills) to GA — beta header skills-2025-10-02 no longer required, including Messages API requests loading Skills via the container parameter.
- ›Adds domain-restriction controls for web_search and web_fetch tools in Claude Managed Agents: set allowed_domains or blocked_domains on the tool's entry in the agent_toolset_20260401 configs array; web_fetch also accepts max_content_tokens and web_search accepts user_location.
- ›Redesigns the session viewer in the Claude Console with a timeline minimap, a transcript grouped by model request, and an Inspector panel covering session details, cost, raw events, per-tool statistics, mounted resources, and per-thread activity.
+1 moreshow less
- ›Both computer_toolset_20260801 and browser_toolset_20260801 are available for Claude Fable 5, Claude Mythos 5, Claude Opus 5, Claude Sonnet 5, and Claude Opus 4.8.
└──▷ BREAKING ON UPGRADE
- !Upgrading an existing computer-use integration to computer_toolset_20260801 changes the request shape and tool handling — existing integrations must follow the migration guide at 'Migrate from computer_20251124'.
- !Files API requests sent without the files-api-2025-04-14 beta header now receive the new response format (which includes expires_at, expires_in_seconds, page/next_page, and ids[]), not the previous format.
August 18, 2026 Aug 18, 2026 · issue 005

Workbench is renamed to Playground, now supporting every Messages API parameter plus built-in templates for code execution and web search.
└──▷ HOW TO FIND IT
Explore and prototype any Messages API parameter interactively, then copy the generated SDK request directly into your integration.
📍In the Claude Console, go to platform.claude.com/playground, select a template (e.g. 'code execution' or 'web search'), adjust parameters, run the prompt, and copy the full SDK request shown in the results panel.
- ›Renames Workbench to Playground at platform.claude.com/playground, now covering every Messages API parameter in an interactive UI.
- ›Adds built-in templates in Playground demonstrating API features including code execution and web search.
- ›Playground surfaces the full SDK request and raw API response for each run, aiding API exploration and integration development.
August 11, 2026 Aug 11, 2026 · issue 005

Compliance API gains local Cowork/Claude Code session transcripts; Claude API now returns workspace ID in response headers.
└──▷ TRY IT
Retrieve the full message transcript of a specific local Cowork or Claude Code session for compliance review.
```
$ curl https://api.anthropic.com/v1/compliance/apps/sessions/local/session_abc123/messages \
  -H 'Authorization: Bearer <compliance_access_key>' \
  -H 'anthropic-version: 2023-06-01'
```
Identify which workspace an API key belongs to by inspecting the response header on any Claude API call.
```
$ curl -I https://api.anthropic.com/v1/messages \
  -H 'x-api-key: <api_key>' \
  -H 'anthropic-version: 2023-06-01' \
  -d '{"model":"claude-opus-4-5","max_tokens":1,"messages":[{"role":"user","content":"hi"}]}' \
  | grep anthropic-workspace-id
```
- ›Adds GET /v1/compliance/apps/sessions/local endpoint to list Cowork and Claude Code sessions running on users' machines across a Claude Enterprise organization (beta).
- ›Adds GET /v1/compliance/apps/sessions/local/{session_id} endpoint to retrieve metadata for a single local session.
- ›Adds GET /v1/compliance/apps/sessions/local/{session_id}/messages endpoint to retrieve the full transcript of a local session.
- ›All three new Compliance API endpoints authenticate with the existing Compliance Access Key and require the read:compliance_user_data scope.
August 7, 2026 Aug 7, 2026 · issue 005

Claude Managed Agents gains session budgets, advisor models, inference geo control, and GitHub-sourced skills
└──▷ USE IT
Give an agent access to a strategic advisor model that the primary thread can consult mid-turn.
json
```
{
  "multiagent": [
    { "type": "advisor", "model": "claude-opus-5" }
  ]
}
```
Pin inference to a specific geography when creating an agent to satisfy data residency requirements.
json
```
{
  "model": {
    "name": "claude-opus-5",
    "inference_geo": "eu"
  }
}
```
- ›Adds spend caps to Claude Managed Agents sessions via a budget field; sessions that hit the cap pause with the budget_reached stop reason and resume when the budget is changed or removed.
August 5, 2026 Aug 5, 2026 · issue 005

Inference hooks (beta) let Enterprise orgs gate every Claude prompt through their own AI security server before inference runs.
- ›Adds inference hooks (beta) for Claude Enterprise organizations: point Claude at your AI security server and every governed prompt across claude.ai, Cowork, and Claude Code is held for an allow/deny verdict before inference proceeds — with signed requests, configurable failure handling, and denials recorded in the compliance Activity Feed.
└──▷ BREAKING ON UPGRADE
- !The claude-opus-4-1-20250805 model has been retired; all API requests to that model now return an error. Upgrade to Claude Opus 5.
August 3, 2026 Aug 3, 2026 · issue 005

Compliance API gains two endpoints to list and retrieve Cowork session transcripts for Claude Enterprise orgs (beta).
└──▷ TRY IT
Retrieve the transcript of a specific Cowork session for audit or eDiscovery review.
```
$ curl -X GET 'https://platform.claude.com/v1/compliance/apps/sessions/remote/<session_id>/messages' \
  -H 'Authorization: Bearer <compliance_access_key>'
```
List all remote Cowork sessions across your Claude Enterprise org to feed into a compliance pipeline.
```
$ curl -X GET 'https://platform.claude.com/v1/compliance/apps/sessions/remote' \
  -H 'Authorization: Bearer <compliance_access_key>'
```
- ›Adds GET /v1/compliance/apps/sessions/remote endpoint to list Cowork sessions started on claude.ai web or mobile, available in beta for Claude Enterprise organizations.
- ›Both endpoints use the existing Compliance Access Key with the read:compliance_user_data scope — no new credentials required.
August 1, 2026 Aug 1, 2026 · issue 005

Claude Dreams research preview now supports Claude Opus 5.
- ›Extends the Dreams research preview to support Claude Opus 5.