Anthropic paused the June 15 billing change - Fable 5 stayed dark all week - SpaceX bought Cursor for $60 billion

Two stories defined the week, and for a newsletter named after usage limits, the one that matters most is the quieter of the two. June 15 was supposed to be the day Claude Agent SDK, claude -p, and GitHub Actions usage left Pro, Max, Team, and Enterprise pools and moved to a separate metered credit pool billed at standard API rates. It didn't. Anthropic paused the change on the morning it was due to take effect, and as of now Agent SDK and headless usage still draw from your normal subscription -- no separate pool, no new bill, no revised date announced. If you spent late May rearranging your CI around the old plan, you can put it back.

The louder story is still unresolved. Claude Fable 5 and Mythos 5 -- pulled June 12 under a US export-control directive over a possible jailbreak -- stayed dark all week, for every user worldwide. Anthropic's international managing director told a Seoul press conference the company is "very confident" the models return "in the coming days," but coming days have a way of becoming next week, and as of this writing both remain offline. Everything else ran normally: Opus 4.8, Sonnet, and Haiku were never touched, and Claude Code shipped six point releases through the saga -- including an auto-mode change that now refuses to run destructive git and terraform commands on your behalf.

And outside Anthropic, the week's biggest move came from a familiar name: SpaceX agreed to buy Cursor's parent, Anysphere, for $60 billion. More on that below.

Anthropic this week

• Anthropic pauses the June 15 Agent SDK billing split - the planned move of Claude Agent SDK, claude -p, and GitHub Actions usage to a separate metered credit pool was pulled back on the morning it was due to take effect. Subscription usage is unchanged for now, and Anthropic has not announced a revised plan or date. The original split was announced May 14
• Fable 5 and Mythos 5 stay suspended; Anthropic "very confident" they return soon - international managing director Chris Ciauri told a Seoul press conference that access should come back "in the coming days." As of this writing both models remain offline worldwide, a week after the June 12 export-control directive. Opus 4.8, Sonnet, and Haiku are unaffected
• Anthropic opens a Seoul office and announces Korean partnerships - the company's third Asia-Pacific office, led by KiYoung Choi, lands mid-ban (June 17) with deployments at LG CNS, Samsung SDS, and Hanwha Solutions, plus Claude for Startups going live in Korea

Claude Code this week

• v2.1.183: auto mode now blocks destructive git commands and terraform operations; model-deprecation warnings print to stderr in print mode; new attribution.sessionUrl setting to omit claude.ai session links; fixes for thinking-disabled 400 errors, WebSearch in subagents, vim-mode cursor glitches, and fullscreen TUI corruption on Windows Terminal
• v2.1.181: /config key=value to set options straight from the prompt; bundled Bun runtime upgraded to 1.4; sandbox.allowAppleEvents opt-in for macOS; fixes for prompt caching on custom base URLs / Foundry and for Write/Edit truncating files on network drives
• v2.1.180 and v2.1.179: long-conversation performance, agent-creation fixes, and mid-stream connection drops now preserving partial responses -- plus the WSL2 mouse-wheel scroll regression chased across both releases
• v2.1.178: removed the TeamCreate / TeamDelete tools (implicit teams now); new Tool(param:value) syntax for permission rules; skills in nested .claude/skills directories load automatically
• v2.1.185: the stream-stall hint ("Waiting for API response -- will retry in ...") now fires after 20 seconds

The broader week

• SpaceX agrees to buy Cursor's parent Anysphere for $60 billion - announced June 16, the all-stock deal has SpaceX exercising a purchase right that xAI struck back in April, valuing the AI-coding startup (roughly $2.6B annualized revenue) at $60B. It is the largest venture-backed startup acquisition on record, expected to close in Q3 pending regulatory approval, and a signal that the Musk orbit now sees agentic coding as core territory against Anthropic and OpenAI
• The Fable 5 ban gave open-weight models an opening - The New Stack (June 18) argues the export-control gap is exactly the kind of moment open models are built to fill, with several responding before Anthropic could restore access. A reminder that a single-vendor outage is a single point of failure for anyone who hasn't kept a fallback