Anthropic Stainless Acquisition, Claude Code 50% Limits, Mythos Cyber Briefing

Anthropic's Power Move: Stainless Acquisition, Claude Code Limits Surge, and the Mythos Cyber Shake-Up

Last week Anthropic went on a spending spree that reshaped the competitive landscape. A $300M+ acquisition, a 50% bump to Claude Code limits, and a cybersecurity model so powerful it's briefing global financial regulators. Here's what landed — and what it means for the people actually building with this stuff.

The Stainless Acquisition: Anthropic Just Locked Down the SDK Layer

Anthropic announced the acquisition of Stainless on May 18. The price tag: north of $300 million.

Stainless isn't a household name, but if you've ever used an Anthropic SDK, you've used Stainless. They built the official TypeScript, Python, Go, Java, and Kotlin SDKs. They also built the MCP server tooling that hundreds of companies depend on.

Here's the real play: Anthropic is now the only company with access to Stainless's SDK generation tools. OpenAI, Google, Cloudflare — they all used Stainless too. That pipeline just got cut.

Caitlin Lesse, Anthropic's Head of Platform Engineering, said it plainly: "The usefulness of an agent depends on what it can connect to."

This isn't a feature release. It's vertical integration. Anthropic just decided that the tooling layer between Claude and every developer ecosystem is theirs.

What this means for you: If you're building SDKs or MCP servers on Stainless's hosted platform, time to evaluate alternatives. The hosted product is being phased out. New registrations, projects, and SDKs are no longer available.

5 Projects I'm Building Right Now — And Why Each One Matters

3 min

No Single Model Wins May 2026 — And That's The Best News For Builders

5 min

Claude Code Limits Up 50% — Stacking With Everything Else

On May 13, Anthropic boosted Claude Code weekly limits by 50% for all Pro, Max, Team, and Enterprise users. The boost runs through July 13, 2026.

But here's the thing most people missed: this stacks with the 5-hour limit doubling from May 6 and the peak-hour removal from April. If you're on Max 20x, you're now looking at roughly 300 weekly hours of Opus 4.7. That's beyond realistic usage even for full-time developers.

The free plan is excluded. This is a retention play — specifically aimed at developers who were migrating to Codex because of its lower token consumption.

The numbers:

Pro ($20/mo): ~60 weekly hours of Sonnet 4.6, ~12 hours of Opus 4.7
Max 5x ($100/mo): ~75 weekly hours of Opus 4.7
Max 20x ($200/mo): ~300 weekly hours of Opus 4.7
Team ($30/user/mo): 50% increase on shared pool
Enterprise: 50% increase on existing contracts, no renegotiation needed

The SpaceX Colossus 1 partnership (300MW+, 220,000+ NVIDIA GPUs) underpins this capacity. Anthropic isn't just throwing money at limits — they're throwing infrastructure.

Mythos: The Cybersecurity Model Too Dangerous to Release

Anthropic's Mythos model — the one they won't publicly release because it's too good at finding and exploiting software vulnerabilities — is now briefing the Financial Stability Board.

The FSB, chaired by Bank of England Governor Andrew Bailey, asked Anthropic to brief them on Mythos's implications for financial system stability. The FSB is drafting "sound practices" for AI in finance, with a public consultation paper coming in June 2026.

Here's the timeline:

Mythos has already found "thousands of critical vulnerabilities" across major operating systems and browsers
~40 organizations have access through Project Glasswing (Amazon, Microsoft, Nvidia, Apple, and others)
The Pentagon is deploying Mythos to find and patch vulnerabilities across the U.S. government
Anthropic just reversed its position: partners can now share Mythos findings with non-Glasswing organizations, security teams, regulators, and the public

The implications are staggering. We're watching a single AI model reshape how the global financial system thinks about cybersecurity.

Claude Code: From CLI to Platform

The Claude Code changelog in May reads like a product roadmap that got compressed into two weeks. The highlights:

Plugin Marketplace Goes Live

Starting with v2.1.108, Claude Code accepts third-party plugins via --plugin-url and --plugin-dir. Plugins can register slash commands, hooks, sub-agents, and skills. The architecture mirrors VS Code extensions, but the execution model is more native.

The official Anthropic marketplace is live. The community-curated list is at awesome-claude-code-plugins.

Worktree Isolation

Multiple Claude Code agents can now run in separate Git worktrees simultaneously. No more edit collisions when you fan out parallel work.

Agent View (Research Preview)

A single list of every Claude Code session — running, blocked, or done. Run claude agents to see everything at once.

Fast Mode Defaults to Opus 4.7

/fast mode now runs on Opus 4.7 instead of 4.6. That's 2.5x faster output with Opus-level reasoning behind it. For multi-file refactoring, this is the default now.

The /radio Easter Egg

On May 8, @ClaudeDevs posted a single line: /radio. It opens Claude FM — a lo-fi radio station Anthropic built to accompany long coding sessions. The terminal is starting to feel like a platform.

Two Patterns Every Claude Code User Should Steal

The 60% Context Rule

Multiple practitioners have converged on the same practice: reset your session when context hits 60%. It's more effective than /compact because you control what survives.

The workflow:

Ask Claude to write current plan and progress to a markdown file
Run /clear to reset
Start fresh with Claude reading that file

/compact lets Claude auto-summarize, which inevitably loses critical context. Manual reset is surgical.

The lessons.md Pattern

Every time Claude makes a mistake, after correcting it, have it add a rule to lessons.md. Over time, Claude literally teaches itself to be better at your project.

# lessons.md
- Import paths always use the @/ alias
- Test files use *.test.ts not *.spec.ts
- API responses always follow { data, error } structure

Keep it separate from CLAUDE.md. Reference it from CLAUDE.md so it auto-loads each session.

SandboxAQ + Anthropic: Drug Discovery Meets Conversational AI

SandboxAQ, chaired by Eric Schmidt, announced a partnership with Anthropic to integrate scientific AI models directly into Claude. Researchers can now access drug discovery and materials science tools through natural language — no specialized computing infrastructure required.

This is the first time a frontier quantitative model has been integrated with a frontier LLM in a conversational interface. If you're in biotech or materials science, this changes the accessibility equation entirely.

The Bigger Picture

Anthropic isn't just building models anymore. They're building infrastructure (Stainless acquisition), expanding capacity (SpaceX partnership, 50% limit increases), entering regulated industries (financial cybersecurity, drug discovery), and scaling enterprise adoption (Hitachi training 100,000 workers).

The coding agent war isn't about who has the best model. It's about who controls the full stack — from SDK generation to compute to deployment to compliance.

Anthropic just made a very loud move in that direction.

What's your take? Is vertical integration the right play, or does this concentration of tooling layers create new risks?

Published May 19, 2026 — Covering Anthropic, Claude Code, and AI engineering developments through May 19, 2026.