This was the biggest week of the year for Anthropic — and possibly the most consequential single release in the AI industry so far in 2026.

What happened: On June 9, Anthropic did something it had never done before. It handed the public a model from its top-secret Mythos tier — the class of models that, until now, only cyber-defense partners and a handful of biology researchers were allowed to touch.

Here’s everything that landed this week.

Claude Fable 5: Mythos Goes Mainstream

Claude Fable 5 is Anthropic’s most capable generally available model — period. It sits above the Opus family in a brand-new tier called Mythos-class.

By the Numbers

  • 80.3% on SWE-Bench Pro (vs 69.2% for Opus 4.8 — the previous flagship)
  • First model to exceed 90% on Hex’s analytical benchmark
  • 1M token context window with 128k max output tokens
  • Always-on adaptive thinking baked into every response

Pricing & Availability

  • $10 per million input tokens, $50 per million output — exactly double Opus pricing
  • Free on Pro, Max, Team, and Enterprise through June 22
  • After June 23: usage credits required
  • Available on: Claude API (claude-fable-5), Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, and the Claude Platform on AWS

How It Works

Fable 5 comes with significant safety infrastructure. On cybersecurity, biology, and chemistry queries — where Anthropic deems the risk highest — the model quietly routes to a less powerful fallback model. This is new territory: shipping a frontier model and a chaperone in the same box.

A new stop_reason: "refusal" API response gives developers programmatic visibility into when the model has deferred a question. Anthropic also shipped an opt-in fallbacks parameter for developers who need custom fallback routing.

Claude Mythos 5

Unlock the same underlying weights. Same model, safety classifiers lifted. Restricted to Project Glasswing partners: US government agencies, AWS, Microsoft, Apple, CrowdStrike, and approximately 200 organizations across 15+ countries. This is the version that finds vulnerabilities in operating systems, not the one that tells you about them.

Claude Code Hits v2.1.175

Claude Code got three significant releases this week:

Dynamic Workflows (Game Changer)

Claude Code can now spin up sub-agents, parallelize work, generate scripts, and validate results across large coding tasks. This transforms it from a coding assistant into a full orchestration layer.

Boris Cherny, head of Claude Code, speaking at an Acquired Unplugged event in San Francisco:

“I don’t prompt Claude anymore. I have loops running that prompt Claude and figuring out what to do. My job is to write loops.”

Multi-Repository Context Windows

Simultaneous analysis across up to 50 linked repositories with a combined context window of 5 million tokens. Early adopters report 40% faster completion times for cross-service feature implementations — this is a direct hit at the microservices refactoring problem that has plagued every AI coding tool.

Other Claude Code Updates

  • Nested sub-agents — agents that spawn agents, recursively
  • Post-session hook & safe mode for troubleshooting
  • \cd — move sessions to a new working directory
  • Fallback models with broader deny-rule glob support
  • Plugin search built into the CLI

Developer Platform: API, SDK & MCP

Three Coordinated SDK Releases

On June 9, all three SDKs (Python, TypeScript, Java) shipped simultaneously with:

  • claude-fable-5 and claude-mythos-5 model IDs
  • Server-side fallbacks-on-refusal + client-side fallbacks middleware
  • ctx.logger for middleware visibility
  • Managed Agents deployment types and environment variable credentials
  • New frontier_llm refusal category

Model Lifecycle

Attention developers: Claude Sonnet 4 (claude-sonnet-4-20250514) and Claude Opus 4 (claude-opus-4-20250514) retire June 15 — three days from now. Migrate to claude-sonnet-4-6 and claude-opus-4-8. Claude Opus 4.1 deprecation follows on August 5.

Apple Developers Get Native Claude Access

A new Swift package lets Apple developers use Foundation Models framework to call Claude directly. Works on iOS 27, iPadOS 27, macOS 27, visionOS 27, and watchOS 27 — handles streaming, tool calls, and structured responses back into SwiftUI.

MCP Ecosystem

The Model Context Protocol consortium published MCP Gateway 1.0 — a standardized proxy layer for enterprise data access. Salesforce, ServiceNow, and SAP have all committed to native MCP Gateway support.

The MCP 2026-07-28 specification introduces a stateless core optimized for HTTP infrastructure with built-in OpenID/OAuth alignment and “Tasks” extensions supporting interactive server-rendered widgets.

Other notable MCP developments:

  • 20+ legal MCP connectors and 12 practice-area plugins for law firms
  • Expensify MCP layer for expense data queries
  • ZoomInfo MCP integration for sales intelligence
  • MCP tunnels (research preview) — agents reach private network services without public endpoints
  • Connector observability dashboard for monitoring MCP performance (public beta)
  • Native Claude support in Chrome, Excel, PowerPoint, and Outlook

Enterprise & Company

Anthropic Files for IPO

In a week of massive product news, Anthropic quietly continued its march toward public markets:

  • Confidential S-1 filed with the SEC on June 1
  • $65B Series H at $965B post-money valuation — led by Altimeter Capital, Dragoneer, Greenoaks, Sequoia, Capital Group, Coatue, and D1 Capital Partners
  • Valuation surpasses OpenAI ($965B vs $852B)

Claude Partner Hub & Services Track

Anthropic formally launched the Claude Partner Network’s Services Track — a three-tier certification system for consulting and implementation partners. The numbers are staggering:

  • 40,000+ firms have applied since the March launch
  • 10,000+ consultants already certified
  • Partners include: DXC Technology (Global Premier), TCS, Infosys, KPMG, Bounteous
  • New MCP connector links the Partner Hub directly with Claude

The 80% Stat That Stopped People

On June 4, Anthropic published a self-issued warning: over 80% of the code merged into production systems in May 2026 was written by Claude — and that number exceeds 90% when scripts and experimental code are included. The company that builds the AI is now eating its own dog food at a level no one predicted.

Enterprise Governance Arrives

The Claude Compliance API launched this week with integrations from Opsin, Linx Security, and Sentra — giving regulated enterprises the visibility and control needed to deploy Claude in production at scale.

Claude Corps

Anthropic launched Claude Corps, a national fellowship program for early-career professionals passionate about extending AI benefits to communities across America.

Ecosystem & Community

Claude’s developer ecosystem continues to accelerate:

  • 3.8 million monthly active developers using Claude Code
  • Claude Design (Anthropic Labs) for visual outputs and prototypes
  • Claude Cowork usage limits doubled through July 5
  • Code with Claude Tokyo — Anthropic’s first international developer conference
  • Claude for Small Business is now live
  • Truecast: open-source tool for managing sub-agents in Claude Code
  • 10 production-ready MCP servers reviewed by InfoWorld
  • Grassroots partnerships with Gates Foundation, Teach For All, CodePath, Allen Institute, HHMI

The Bottom Line

This was the week Anthropic stopped running second.

Fable 5 is a genuine generational leap — the first time a company has given the public access to its most capable internal model tier. The safety architecture (chaperone model, refusal signals, fallback chain) sets a precedent for responsible frontier deployment. And the combination of IPO filing, partner ecosystem, and developer tooling suggests Anthropic is playing for dominance, not just survival.

For developers: Migrate off Sonnet 4 and Opus 4 before Tuesday. Start experimenting with Fable 5 before June 22 while it’s free. And pay attention to MCP Gateway 1.0 — the enterprise adoption signal is stronger than most people realize.

For enterprises: The compliance API, self-hosted sandboxes, and MCP tunnels make Claude deployable in regulated environments. The partner ecosystem (DXC, TCS, Infosys, KPMG) gives you certified implementers. The IPO filing means long-term product stability. It’s time to evaluate Claude seriously.

This is the Claude Weekly by Big Hat Group — covering the Anthropic and Claude ecosystem every week.