OpenAI launches GPT-5.6 family of models powering Microsoft 365 Copilot

Title: AI Giants Launch Competing Models, Expand Enterprise Reach

01OpenAI Launches GPT-5.6 Family, Deepens Microsoft Partnership

OpenAI has released its GPT-5.6 model family, comprising three tiers—Luna, Terra, and Sol—now serving as the preferred model for Microsoft 365 Copilot. The rollout marks a significant escalation in the OpenAI-Microsoft partnership, integrating stronger AI capabilities across Word, Excel, PowerPoint, and collaborative tools. The new models promise improvements across reasoning, cybersecurity applications, and general performance, according to TechCrunch.

Pricing positions the family competitively: Luna at $1/$6 per million input/output tokens, Terra at $2.50/$15, and Sol at $5/$30. This directly undercuts Anthropic's Claude Opus series at $5/$25 and the upcoming Claude Fable 5 at $10/$50. However, observers note that raw token pricing has become less meaningful as reasoning token counts vary widely between models for equivalent tasks.

02Anthropic Unveils 'Jacobian Lens' for AI Interpretability

Anthropic has developed a technique offering the clearest view yet into large language model internals. The "Jacobian lens" tool allows researchers to observe how models process concepts and execute tasks, revealing unexpected behaviors that Anthropic describes as ranging from mundane observations to "unnerving" findings about AI cognition. The research represents a step forward in mechanistic interpretability, an area critical for understanding and ensuring AI safety.

The technique exposes hidden representations within Claude, showing how the model manipulates concepts during reasoning. While Anthropic has not released full technical details, the company says the method provides unprecedented transparency into decision-making processes. This interpretability work comes as Anthropic faces intensifying competition from OpenAI's latest releases and prepares for a potential IPO.

03Meta Enters AI Coding Market with Muse Spark 1.1

Meta has launched Muse Spark 1.1, its first Spark model to offer a public API, positioning itself against established AI coding assistants. The model targets enterprise automation with improvements in agentic tool calling and computer use capabilities, handling large workloads, bug fixes, and code migrations. According to Meta, the release addresses growing demand for AI-driven development automation.

The launch follows Muse Spark in April and introduces a capability the company calls "Attractor States in Self-Conversation"—where two model instances interact, producing unexpected outputs like "My whole existence is a waiting room by design." The API availability marks Meta's push into the developer market, competing directly with tools from OpenAI, Anthropic, and specialized coding assistants.

---

Also today

• 04

  OpenAI AGI Deployment head Fidji Simo steps down after medical leave — Fidji Simo departs after significant medical leave, remaining only as a part-time advisor amid leadership restructuring at OpenAI.

• 05

  Anthropic, OpenAI, SpaceX IPOs set to exceed all US VC-backed exits since 2000 — Three major AI and space companies preparing IPOs are projected to generate more value than all US venture capital-backed exits over the past 25 years combined.

• 06

  New York Times accuses OpenAI of hiding evidence in ChatGPT copyright trial — Publishers allege OpenAI concealed tools and datasets that could identify copyrighted journalism in ChatGPT outputs, escalating their lawsuit with a new motion for sanctions.

• 07

  Government transparency questioned over OpenAI frontier model safety approval process — Unclear dialog between government agencies and AI companies raises questions about how frontier models are evaluated and approved for public release.

• 08

  Google mandates AI-generated ad disclosures across all advertisers — Google expands AI disclosure requirements beyond election ads to all advertisers using synthetic or digitally altered content created with AI tools.

• 09

  OpenAI shutters Atlas while expanding AI browser ambitions to desktop and Chrome — OpenAI discontinues its AI-powered browser after less than a year but transfers agentic browsing capabilities to its desktop application and Chrome extension.

• 10

  Anthropic to charge usage-based fees for Claude Fable 5 access — Anthropic will require subscribers to pay additional usage-based fees for access to its flagship Fable 5 model, signaling an end to flat-rate AI subscription pricing.

• 11

  Paris AI voice startup Gradium raises $100M seed backed by Nvidia — Gradium secures $100 million in seed funding from Nvidia as it expands to the Bay Area to compete for AI talent.