CLAUDE·news
PTEN
№ 0592026-06-10

Claude Report — 2026-06-10

🚀 Anthropic Official

Claude Fable 5 & Claude Mythos 5 — First Public Mythos-Class Models

Launched June 9, 2026. Claude Fable 5 (claude-fable-5) is Anthropic's most capable generally available model; Claude Mythos 5 (claude-mythos-5) is the same underlying model with safety classifiers lifted, restricted to Project Glasswing participants.

  • Specs: 1M-token context window, 128k max output tokens; priced at $10 / $50 per M input/output tokens (2× Opus 4.8)
  • Adaptive thinking always-on: thinking: {"type": "disabled"} not supported; use effort param to control depth; raw chain of thought never returned — set display: "summarized" to receive readable summaries; pass thinking blocks back unchanged in multi-turn conversations
  • Safety classifiers: block cyber / bio / chem / distillation requests; blocked requests return stop_reason: "refusal" (HTTP 200, not an error) with which classifier fired; opt-in fallbacks parameter (beta on Claude API and Claude Platform on AWS) auto-retries on another model, with fallback credit covering prompt-cache cost on the switch
  • Tokenizer note: same tokenizer as Opus 4.7+ — same text produces ~30% more tokens vs pre-Opus-4.7 models; re-measure context budgets via token counting API with model: "claude-fable-5" before migrating
  • Data retention: 30-day retention required for both models (no ZDR option available)
  • Availability: Claude API, Amazon Bedrock, Vertex AI, Microsoft Foundry; free for Pro/Max/Team/Enterprise through June 22, then requires usage credits

Claude Code v2.1.170 — Fable 5 Access + VS Code Transcript Fix

Released June 9, 17:23 UTC. Unlocks Claude Fable 5 access in Claude Code (update required). Fixes sessions not saving transcripts and not appearing in --resume when launched from the VS Code integrated terminal or any shell that inherited Claude Code environment variables.

Anthropic SDK — TypeScript v0.103.0–v0.104.1 & Python v0.108.0–v0.109.1

Three coordinated releases on June 9 ship Fable 5 support across both SDKs: claude-fable-5 and claude-mythos-5 model IDs, server-side fallbacks-on-refusal wiring, client-side fallbacks middleware for platforms without server-side fallback support, ctx.logger added to middleware, Managed Agents deployment types and environment variable credentials, and a new frontier_llm refusal category. Python also ships v0.107.1 (June 7) fixing Foundry API-key auth — the x-api-key header was not sent correctly for API-key auth on Foundry.

🔌 Claude Code Plugins

Plugin cache unavailable — top-10 tracking skipped this run.

🛠️ Skills

None of the top 5 updated in window.

Unchanged in window: glebis/claude-skills, anthropics/skills, anthropics/knowledge-work-plugins, Piebald-AI/claude-code-system-prompts, ComposioHQ/awesome-claude-skills

🤖 Agents & Subagents

claude-agent-sdk-typescript v0.3.170 — Fable 5 Model Types

Released June 9, 17:23 UTC at parity with Claude Code v2.1.170. Adds the claude-fable-5 model type and a fable shorthand alias to SDK model types so TypeScript agent code can target Fable 5 without raw string literals.

Unchanged in window: nicobailon/pi-subagents, Chachamaru127/claude-code-harness

🔗 MCPs & Integrations

GitHub Copilot — Claude Fable 5 Generally Available

Published June 9, 2026. Adds Fable 5 to Copilot Pro+, Max, Business, and Enterprise across VS Code, Visual Studio, JetBrains, Xcode, Eclipse, Copilot CLI, GitHub Mobile, and GitHub.com. Unlike other Claude models in Copilot (which offer zero data retention), Fable 5 requires 30-day data retention to operate safety classifiers; Enterprise/Business admins must explicitly enable the Fable 5 policy in Copilot settings — it defaults to disabled.

modelcontextprotocol/servers — Everything Server Infinite Loop Fix

Committed June 7, 2026. Fixes an infinite loop in the everything MCP reference server that triggered during URL elicitation on HTTP transport error paths, preventing the server from hanging indefinitely on certain failure modes.

Unchanged in window: modelcontextprotocol/typescript-sdk; not verified in window: github/github-mcp-server, awslabs/mcp, portainer/portainer-mcp

💡 Community — Workflows & Ideas

HN: "Claude Fable 5" — 2,256 pts, 1,744 comments (June 9)

Community testing surfaces that Fable 5 achieves equivalent output quality in roughly half the token count vs Opus 4.8, partially offsetting the 2× price increase. Top friction reported: safety classifiers blocking legitimate biological research and cybersecurity tooling, leading several teams to route those workloads back to Opus 4.8.

📰 Quick Mentions

Relevant links that don't warrant a section of their own:

← back to archiveend of edition № 059