AI Landscape Snapshot — Week 24

Claude Fable 5: Anthropic's New #1

Anthropic released Claude Fable 5 and Claude Mythos 5 on June 9. Fable 5 immediately took the top spot on the Artificial Analysis LLM Leaderboard with an Intelligence Index score of 65.

The core story: Fable 5 and Mythos 5 share the same underlying weights. Mythos 5 is the unrestricted version, available only to roughly 200 organizations through Project Glasswing. Fable 5 is the public release with safety classifiers that route cybersecurity, biology, chemistry, and model distillation queries to Claude Opus 4.8. These classifiers trigger in fewer than 5% of sessions, and flagged queries are not billed at Fable 5 rates.

The benchmark numbers tell a clear story. On SWE-Bench Pro, Fable 5 scored 80.3% versus 69.2% for Opus 4.8 and 58.6% for GPT-5.5. On SWE-Bench Verified, it hit 95.0%. On Humanity's Last Exam with tools, it scored 64.5% compared to 52.2% for GPT-5.5. Cursor CEO Michael Truell noted that Fable 5 is state-of-the-art on CursorBench, opening up long-horizon problems that were previously out of reach.

Specs: 1M token context window, 128K max output, $10/$50 per million tokens. That's double Opus 4.8's pricing ($5/$25). One important caveat: Anthropic now requires 30-day data retention for all traffic on Mythos-class models, including on third-party surfaces.

Fable 5 is available on the Claude API, AWS Bedrock, Vertex AI, Microsoft Foundry, and GitHub Copilot.

Apple Rebuilds Siri with Gemini

Apple opened WWDC 2026 on June 8 with the announcement it had been delaying for two years: a completely rebuilt Siri, now branded "Siri AI." The assistant is powered by Google's Gemini models under a deal worth roughly $1 billion per year.

The new architecture runs across three layers: on-device Apple Foundation Models, Apple's Private Cloud Compute, and Gemini for complex queries. iOS 27 introduces Extensions that let users select a third-party AI model as their default assistant — a meaningful opening for developers.

The EU dimension matters: the European Commission confirmed on June 9 that Siri AI will not launch in Europe, with Apple and regulators publicly disagreeing on who is blocking whom.

Also announced: iOS 27, macOS 27 Golden Gate, and performance improvements including 30% faster app launches and 70% faster photo previews.

SpaceX IPO: Largest in History

SpaceX (SPCX) opened trading on Nasdaq on June 12 at $150 per share, above the $135 IPO price, and closed at $161 — a 19% first-day gain. The company raised $75 billion at an implied valuation of approximately $1.77 trillion, making it the largest IPO ever.

The AI angle: SpaceX includes the xAI division (Grok), merged in February 2026. xAI lost $2.4 billion in Q1 2026 alone while spending $7.7 billion in capex. The profitable Starlink segment ($11.4B revenue, $4.4B operating income in 2025) carries the valuation.

Both Frontier AI Labs File for IPO

OpenAI acknowledged a confidential S-1 filing on June 8, one week after Anthropic filed its own on June 1. Both companies are targeting 2026 listings.

The financial profiles are strikingly different. Anthropic's annualized revenue reached $44-47 billion as of May 2026, and the company is on track for its first operating profit (~$559M) in Q2 2026. OpenAI has over $20 billion in ARR but projects $14 billion in losses for 2026. Both are seeking valuations near or above $1 trillion.

For practitioners, the near-term implication is concrete: public-market scrutiny tends to push loss-making providers toward profitability, which means the era of below-cost API pricing may be closing.

EU AI Act: 49 Days Out

The bulk of the EU AI Act takes effect on August 2, 2026. The GPAI Code of Practice has been signed by most major providers — Anthropic, Google, IBM, Microsoft, Mistral, OpenAI, and Cohere — but Meta has not signed and xAI signed only the Safety and Security chapter. Organizations deploying AI in the EU should be completing compliance assessments now, not at the deadline.

Other Notable Developments

• GitHub Copilot switched from flat subscription pricing to token-based billing on June 1
• OpenAI released GPT-Rosalind (life sciences model update) on June 4 and granted the EU access to GPT-5.5-Cyber
• Anthropic launched the Claude Partner Hub and a $100M partner program for enterprises
• Anthropic expanded Project Glasswing to hundreds of organizations across 15 countries covering critical infrastructure sectors
• OpenAI is adjusting its 2028 autonomous AI researcher milestone to a "tandem" human-AI framing rather than full autonomy

Current Model Landscape

The top of the leaderboard as of this week:

What This Means for Practitioners

If you're building coding agents, Fable 5's SWE-Bench Pro lead is significant — an 11-point jump over its predecessor is not a typical incremental release. The 30-day data retention requirement and safety classifier fallbacks are the trade-offs to evaluate.

If you're choosing a model stack, the pricing spread from Qwen 3.5 at fractions of a cent to Fable 5 at $50/MTok output means model routing is now a first-class architectural concern, not an optimization.

If you serve EU users, the August 2 deadline is real and approaching fast. The Siri AI block in Europe shows that even Apple-scale companies are hitting regulatory walls. Check whether your AI deployments need a Fundamental Rights Impact Assessment.

If you depend on AI API pricing, both Anthropic and OpenAI heading toward IPOs signals that current pricing may not last. Lock in contracts where possible and build provider portability into your architecture.

Assess where you stand with the AISA assessment. Understand the competencies that matter with our AI skills rubric. And explore how these developments map to specific roles.