AI Landscape Snapshot — Week 25

The Fable 5 Ban Enters Day Nine

The biggest story in AI this week isn't a launch — it's an absence. Anthropic's Claude Fable 5 and Claude Mythos 5, the highest-scoring models on the Artificial Analysis Intelligence Index, have been offline since June 12 following a US government export control directive. As of June 21, the claude-fable-5 API endpoint still returns errors.

For context: Fable 5 launched June 9 as Anthropic's first publicly available Mythos-class model — the same underlying weights as the restricted Mythos 5, but with a safety classifier layer that routes flagged requests (cybersecurity, biology, chemistry) to Claude Opus 4.8. It scored 80.3% on SWE-Bench Pro, priced at $10/$50 per million tokens with a 1M token context window.

Anthropicʼs Managing Director of International, Chris Ciauri, said at the company's Seoul office opening on June 17-18 that he is confident the models will return "in the coming days." President Trump separately characterized the negotiations as "going fine."

Practitioners who relied on Fable 5 are routing to Claude Opus 4.8 ($5/$25, still available) or evaluating alternatives — which brings us to the next story.

GLM-5.2: The Open-Weights Model Filling the Gap

Z.ai (formerly Zhipu AI) released GLM-5.2 on June 16 under an MIT license with no usage restrictions. It's a 744B-parameter mixture-of-experts model (40B active) with a 1M token context window — up from 200K in GLM-5.1.

The timing isn't subtle. The same week the US demonstrated it can pull frontier models offline overnight, a Chinese lab released what Artificial Analysis ranks as the strongest open-weights model currently available. GLM-5.2 scores 91.2% on GPQA Diamond and 62.1% on SWE-bench Pro. Via hosted APIs, it runs at roughly $1.40/$4.40 per million tokens — a fraction of GPT-5.5's $5/$30 or Opus 4.8's $5/$25.

It's natively compatible with Claude Code, Cline, Roo Code, and Ollama. For teams evaluating self-hostable alternatives, it's the current benchmark.

Noam Shazeer Leaves Google for OpenAI

On June 18, Noam Shazeer announced he is leaving Google to join OpenAI. Shazeer co-authored the 2017 "Attention Is All You Need" paper that introduced the Transformer architecture. He was VP of engineering at Google and co-lead of Gemini. Google paid roughly $2.7 billion to bring him back from Character.AI in 2024.

He will reportedly serve as Lead for Architecture Research at OpenAI. In the same week, OpenAI also hired Dean Ball, a former White House AI policy official, to lead a new Strategic Futures team — both moves ahead of OpenAI's anticipated IPO.

SpaceX Acquires Cursor for $60 Billion

SpaceX filed its $60B all-stock acquisition of Cursor (Anysphere) with the SEC on June 16, four days after its record Nasdaq IPO. Cursor has approximately $4 billion in annualized revenue and claims deployment across 64% of the Fortune 500.

The deal matters to practitioners because Cursor has historically relied on Claude and GPT models. With SpaceX/xAI as its owner — and xAI a direct competitor — there are real questions about whether third-party model access inside Cursor will be progressively restricted. A joint AI coding model trained on xAI's Colossus infrastructure is reportedly in development.

Closing is expected Q3 2026 pending regulatory approval.

Model Leaderboard Snapshot

Tools and Platform Updates

• Gemini CLI deprecated June 18. Replaced by Antigravity CLI, a Go-based terminal tool sharing Antigravity 2.0's agent harness. If you were on Gemini CLI, migrate now.
• Claude Design gained two-way Claude Code integration on June 18 — a /design terminal command and design-to-code handoff.
• OpenAI Partner Network launched with $150M backing, aiming to certify 300,000 consultants by end of 2026.
• FERC issued show-cause orders to six US regional grid operators on June 18 to accelerate AI data center interconnection.

Regulatory Watch

• Colorado AI Act: Enforcement begins June 30 — nine days from now. Companies that delayed compliance expecting federal preemption are running out of time.
• EU AI Act: Bulk application starts August 2, 2026 — 42 days out.

What This Means for Practitioners

If you use Claude Fable 5: It's offline. Route to Opus 4.8 or evaluate GLM-5.2 for cost-sensitive workloads. Monitor Anthropic's restoration timeline.

If you use Cursor: The SpaceX acquisition hasn't changed anything yet, but start thinking about model access dependencies. The precedent from the Windsurf/OpenAI situation (where Anthropic cut Claude access during acquisition) is instructive.

If you're choosing a coding tool: Claude Code, Codex, Cursor, and Antigravity 2.0 have converged on similar agentic patterns. Teams increasingly use two or more in parallel — architecture in one, implementation in another.

If you ship AI products: The Colorado AI Act is nine days from enforcement. The EU AI Act is six weeks out. Both are real compliance deadlines, not theoretical ones.

Take the AISA assessment to benchmark your current AI skills against this shifting landscape. The AI skills rubric maps these developments to the competencies that matter for developers and other technical roles.