{"id":"org_HXldkKAl-bZUWW6lNnELP","slug":"anthropic","name":"Anthropic","domain":"anthropic.com","description":null,"category":null,"tags":[],"sourceCount":5,"releaseCount":272,"releasesLast30Days":34,"avgReleasesPerWeek":8.6,"lastFetchedAt":"2026-04-17T21:23:20.592Z","trackingSince":"2024-09-19T00:00:00.000Z","aliases":["claude.ai","claude.com"],"accounts":[{"platform":"github","handle":"anthropics"}],"products":[{"id":"prod_4dRP6KdnUG6b8tcZdtiHV","slug":"claude","name":"Claude","url":null,"description":"Claude AI assistant","sourceCount":1},{"id":"prod_QMqJRAH7ip4wF-kRw5Q8f","slug":"claude-code","name":"Claude Code","url":null,"description":"AI coding agent for the terminal and IDE","sourceCount":1},{"id":"prod_BE0qCzI8J0bhXsJUbEhdR","slug":"claude-platform","name":"Claude Platform","url":null,"description":"Developer platform and API","sourceCount":1}],"sources":[{"id":"src_awwhhghp-mRowyjfKMDKt","slug":"anthropic-engineering-blog","name":"Anthropic Engineering Blog","type":"scrape","url":"https://www.anthropic.com/engineering","isPrimary":false,"isHidden":false,"fetchPriority":"low","releaseCount":3,"latestVersion":null,"latestDate":"2025-11-24T00:00:00.000Z","latestAddedAt":"2026-04-10T14:17:30.543Z","productSlug":null,"productName":null},{"id":"src__1wW8sp0gv9nbxCNdXeXW","slug":"anthropic-news","name":"Anthropic News","type":"scrape","url":"https://www.anthropic.com/news","isPrimary":false,"isHidden":false,"fetchPriority":"low","releaseCount":2,"latestVersion":"Opus 4.7","latestDate":"2026-04-17T00:00:00.000Z","latestAddedAt":"2026-04-17T21:23:20.504Z","productSlug":null,"productName":null},{"id":"src_ZDMjMmrcSRlhsc0glSCYV","slug":"claude","name":"Claude","type":"scrape","url":"https://support.claude.com/en/articles/12138966-release-notes","isPrimary":false,"isHidden":false,"fetchPriority":"normal","releaseCount":44,"latestVersion":null,"latestDate":"2026-04-16T00:00:00.000Z","latestAddedAt":"2026-04-16T14:58:06.784Z","productSlug":"claude","productName":"Claude"},{"id":"src_uDHhSwgE7ufu05-FTCFtr","slug":"claude-code","name":"Claude Code","type":"scrape","url":"https://code.claude.com/docs/en/changelog","isPrimary":false,"isHidden":false,"fetchPriority":"normal","releaseCount":206,"latestVersion":"2.1.112","latestDate":"2026-04-16T00:00:00.000Z","latestAddedAt":"2026-04-17T21:07:44.751Z","productSlug":"claude-code","productName":"Claude Code"},{"id":"src_reidosAHay-vJROI8wSJB","slug":"claude-platform","name":"Claude Platform","type":"scrape","url":"https://platform.claude.com/docs/en/release-notes/overview","isPrimary":false,"isHidden":false,"fetchPriority":"normal","releaseCount":17,"latestVersion":"Opus 4.7","latestDate":"2026-04-16T00:00:00.000Z","latestAddedAt":"2026-04-16T14:57:42.805Z","productSlug":"claude-platform","productName":"Claude Platform"}],"overview":{"scope":"org","content":"**Opus 4.7 launched as flagship reasoning model with breaking API changes.** Opus 4.7 replaces Opus 4.6 as the most capable generally available model for complex reasoning and agentic coding tasks, maintaining the same $5/$25 per MTok pricing. The upgrade includes vision improvements and API breaking changes — review the migration guide before upgrading. Simultaneously, Claude Sonnet 4 and Opus 4 entered deprecation with retirement scheduled for June 15, 2026; migrate to Sonnet 4.6 and Opus 4.6 respectively.\n\n**Extended thinking display field and interactive effort tuning shipped.** The new `display` field for extended thinking lets you omit thinking content from responses for faster streaming while preserving the signature for multi-turn continuity — billing unchanged. The `/effort` command now opens an interactive slider for tuning speed vs. intelligence on Opus 4.7, with `xhigh` sitting between `high` and `max`. Pro users see improved cost breakdowns by model and cache-hit rates, plus cache-expiration hints with token counts in a new format (`1.5m` for millions).\n\n**1M token context window now generally available; media limit raised to 600.** The 1M token context window is generally available for Opus 4.6 and Sonnet 4.6 at standard pricing with no beta header required. Media limit increased from 100 to 600 images/PDF pages per request with 1M context. The Models API now returns `max_input_tokens`, `max_tokens`, and a `capabilities` object for programmatic feature discovery. Max tokens cap raised to 300k on the Message Batches API (include `output-300k-2026-03-24` beta header).\n\n**Managed Agents and `ant` CLI launched for autonomous operation.** The Managed Agents API (beta, requires `managed-agents-2026-04-01` header) provides fully managed agent harness with secure sandboxing, built-in tools, and server-sent event streaming. The new `ant` CLI offers rapid Claude API interaction, native Claude Code integration, and YAML-based resource versioning.\n\n**Computer use research preview available for Pro/Max without setup.** Claude can click, type, navigate, and perform tasks directly on-screen — no client installation required. Deepens Cowork Dispatch usefulness by allowing Claude to act on your behalf. Claude Code added computer-use support with the `/tui` command for flicker-free rendering and push notification tool integration for mobile.\n\n**Advisor tool in public beta for high-quality long-horizon tasks.** Pair a faster executor model with a higher-intelligence advisor that provides strategic guidance mid-generation. Achieves near-advisor-solo quality while bulk token generation happens at executor rates. Include the `advisor-tool-2026-03-01` beta header.\n\n**Claude Cowork reached general availability with enterprise controls.** Available on macOS and Windows via Claude Desktop, Cowork gains Analytics API access for engagement metrics, usage analytics, and OpenTelemetry support. Enterprise admins control role-based access via custom roles and group management (manual or SCIM), restricting Claude capabilities by department.\n\n**Interactive content and office integrations expanded.** Claude for iOS and Android now render fully interactive apps with live charts, sketches, and shareable assets. Claude's core model generates custom charts, diagrams, and visualizations inline. Excel and PowerPoint add-ins share full conversation context across applications and support skills via LLM gateways (Bedrock, Vertex","releaseCount":109,"lastContributingReleaseAt":"2026-04-16T00:00:00.000Z","generatedAt":"2026-04-08T02:04:27.602Z","updatedAt":"2026-04-17T21:08:18.156Z"},"knowledgePage":{"scope":"org","content":"**Opus 4.7 launched as flagship reasoning model with breaking API changes.** Opus 4.7 replaces Opus 4.6 as the most capable generally available model for complex reasoning and agentic coding tasks, maintaining the same $5/$25 per MTok pricing. The upgrade includes vision improvements and API breaking changes — review the migration guide before upgrading. Simultaneously, Claude Sonnet 4 and Opus 4 entered deprecation with retirement scheduled for June 15, 2026; migrate to Sonnet 4.6 and Opus 4.6 respectively.\n\n**Extended thinking display field and interactive effort tuning shipped.** The new `display` field for extended thinking lets you omit thinking content from responses for faster streaming while preserving the signature for multi-turn continuity — billing unchanged. The `/effort` command now opens an interactive slider for tuning speed vs. intelligence on Opus 4.7, with `xhigh` sitting between `high` and `max`. Pro users see improved cost breakdowns by model and cache-hit rates, plus cache-expiration hints with token counts in a new format (`1.5m` for millions).\n\n**1M token context window now generally available; media limit raised to 600.** The 1M token context window is generally available for Opus 4.6 and Sonnet 4.6 at standard pricing with no beta header required. Media limit increased from 100 to 600 images/PDF pages per request with 1M context. The Models API now returns `max_input_tokens`, `max_tokens`, and a `capabilities` object for programmatic feature discovery. Max tokens cap raised to 300k on the Message Batches API (include `output-300k-2026-03-24` beta header).\n\n**Managed Agents and `ant` CLI launched for autonomous operation.** The Managed Agents API (beta, requires `managed-agents-2026-04-01` header) provides fully managed agent harness with secure sandboxing, built-in tools, and server-sent event streaming. The new `ant` CLI offers rapid Claude API interaction, native Claude Code integration, and YAML-based resource versioning.\n\n**Computer use research preview available for Pro/Max without setup.** Claude can click, type, navigate, and perform tasks directly on-screen — no client installation required. Deepens Cowork Dispatch usefulness by allowing Claude to act on your behalf. Claude Code added computer-use support with the `/tui` command for flicker-free rendering and push notification tool integration for mobile.\n\n**Advisor tool in public beta for high-quality long-horizon tasks.** Pair a faster executor model with a higher-intelligence advisor that provides strategic guidance mid-generation. Achieves near-advisor-solo quality while bulk token generation happens at executor rates. Include the `advisor-tool-2026-03-01` beta header.\n\n**Claude Cowork reached general availability with enterprise controls.** Available on macOS and Windows via Claude Desktop, Cowork gains Analytics API access for engagement metrics, usage analytics, and OpenTelemetry support. Enterprise admins control role-based access via custom roles and group management (manual or SCIM), restricting Claude capabilities by department.\n\n**Interactive content and office integrations expanded.** Claude for iOS and Android now render fully interactive apps with live charts, sketches, and shareable assets. Claude's core model generates custom charts, diagrams, and visualizations inline. Excel and PowerPoint add-ins share full conversation context across applications and support skills via LLM gateways (Bedrock, Vertex","releaseCount":109,"lastContributingReleaseAt":"2026-04-16T00:00:00.000Z","generatedAt":"2026-04-08T02:04:27.602Z","updatedAt":"2026-04-17T21:08:18.156Z"},"playbook":{"scope":"playbook","content":"# Anthropic — Playbook\n\n> Agent reference for fetching and maintaining **Anthropic** (`anthropic`) changelog sources.\n\n**5** active sources · **3** products: Claude, Claude Code, Claude Platform · domain: anthropic.com\n\n## Sources\n\n| Name | ID | Type | URL | Product | Last Fetched |\n|------|-----|------|-----|---------|--------------|\n| Claude Code | `src_uDHhSwgE7ufu05-FTCFtr` | scrape | https://code.claude.com/docs/en/changelog | Claude Code | Apr 17 |\n| Claude | `src_ZDMjMmrcSRlhsc0glSCYV` | scrape | https://support.claude.com/en/articles/12138966-release-notes | Claude | Apr 16 |\n| Claude Platform | `src_reidosAHay-vJROI8wSJB` | scrape | https://platform.claude.com/docs/en/release-notes/overview | Claude Platform | Apr 16 |\n| Anthropic Engineering Blog | `src_awwhhghp-mRowyjfKMDKt` | scrape · low | https://www.anthropic.com/engineering | — | Apr 10 |\n| Anthropic News | `src__1wW8sp0gv9nbxCNdXeXW` | scrape · low | https://www.anthropic.com/news | — | Apr 17 |\n\n## Parse Instructions\n\n> To update, use `edit_source` with metadata — do not edit the playbook header.\n\n**Claude** (`claude`): This changelog uses dates as section headers (e.g. \"March 25, 2026\"). These are NOT version strings. Set version to null for every release. Use the date as publishedAt only.\n\n**Anthropic Engineering Blog** (`anthropic-engineering-blog`): This is an engineering blog mixing product announcements with educational content. ONLY extract posts that announce a NEW product, feature, tool, service, or capability that users/developers can now use. Signals: titles containing \"Introducing\", \"launching\", or describing something new. SKIP: best practices guides, benchmark analyses, eval methodology, postmortems, technical deep-dives, and educational content — even if they reference a product. If the post is primarily educational (\"how to\", \"building with\", \"designing\", \"demystifying\"), skip it. When in doubt, skip.\n\n**Anthropic News** (`anthropic-news`): Corporate news page — mixes product launches with partnerships, policy, and funding announcements. HIGH NOISE but occasionally carries critical launches (e.g. Anthropic Labs, Claude Design, new consumer products) that don't appear on any other source. Every post MUST be evaluated individually.  INCLUDE when the post announces something users or developers can now use or try: - New model launches (Opus, Sonnet, Haiku, and variants) - New products or apps (e.g. Anthropic Labs, Claude Design, Claude for X) - Major features or capabilities users can access today - New platform services, tools, or SDKs - Significant expansions of existing products (e.g. GA of a tier, new region)  SKIP when the post is primarily about: - Partnerships, customer stories, government MOUs, policy statements - Office openings, hiring announcements, exec moves - Funding, investment news, acquisitions of other companies - Research publications, safety reports, evaluation methodologies - Opinion pieces, industry commentary, thought leadership  When in doubt: if the post describes something a user or developer can USE or TRY, include it. Noise is acceptable; missing a real launch is not.\n\n## Agent Notes\n\n### Fetch instructions\n\n**Claude Code** (`claude-code`) — Static HTML, no rendering needed. Parse `<h2>` elements as version boundaries — each is a semver string (e.g. \"Version 2.1.98\"). Expect 2–4 new releases per week. All content is relevant; no filtering needed.\n\n**Claude** (`claude`) — Set version=null for every release. Section headers are dates (e.g. \"April 9, 2026\"), not versions. `parseInstructions` handle this automatically. Weekly cadence. All entries are real product changes.\n\n**Claude Platform** (`claude-platform`) — Date-grouped entries, no version strings. All content is relevant. 1–3 releases per week when active. After extraction, check for doubled relative links in stored URLs (see Traps).\n\n**Engineering Blog** (`anthropic-engineering-blog`) — Low priority. Only fetch when looking for launch announcements specifically. ~95% of posts are educational/research — `parseInstructions` filter aggressively. Yields ~1–3 relevant items per quarter. Index page shows cards only; crawl mode untested.\n\n**Anthropic News** (`anthropic-news`) — Low priority, HIGH NOISE, intentionally active. This is the ONLY surface for certain launches (Anthropic Labs, Claude Design, new consumer products) — they don't hit any other changelog. No feed available, so every fetch re-evaluates /news from scratch. Expect to reject ~95% of candidates as partnerships, policy, funding, or research posts. The agent MUST evaluate each post individually against `parseInstructions`: the rule is \"when in doubt, include\" — missing a real launch is worse than indexing a customer story.\n\n### Traps\n\n- **Doubled paths on Claude Platform**: Relative doc links get prefixed with the source URL during extraction, producing URLs like `platform.claude.com/docs/en/release-notes/overview/docs/en/...`. Non-blocking but noisy in stored content.\n- **Blog false positives**: If engineering blog starts yielding how-to guides or benchmarks, the parseInstructions need tightening — those are educational, not releases.\n- **News noise floor**: `anthropic-news` is expected to pull in many junk candidates and filter most out. That's the cost of not missing launches like Anthropic Labs. If partnerships / policy / funding posts start getting through, tighten `parseInstructions` — do NOT disable the source. Previously hidden as low-signal; re-enabled 2026-04-17 after missing the Claude Design / Anthropic Labs launch.\n\n### Coverage\n\nThe four active sources cover Anthropic's full release surface:\n\n- **Changelogs** (Claude, Claude Code, Claude Platform) — versioned product updates, shipped weekly\n- **Engineering Blog** — occasional launch context that predates changelog entries (~1–3/quarter)\n- **Anthropic News** — catch-all for consumer-facing launches (Anthropic Labs, Claude Design) that don't appear on the changelogs\n\nNo known gaps.\n","updatedAt":"2026-04-18T12:25:14.553Z"}}