Two frontier models shipped this week behind government gates, and an orchestration approach kept making the case that the model isn’t always the product.
- OpenAI previews GPT-5.6 Sol, Terra, and Luna — On June 26 OpenAI opened a limited preview of three GPT-5.6 models, with access initially capped at roughly 20 partner organizations at the U.S. government’s request.
- U.S. clears Anthropic to release Claude Mythos 5 to select institutions — On June 26 the government revised its earlier export block, letting Anthropic distribute the cybersecurity-focused Mythos model to over 100 vetted U.S. companies and agencies.
- Sakana AI launches Fugu and Fugu Ultra — On June 22 Sakana released an orchestration system that routes tasks across a swappable pool of frontier LLMs behind one API, claiming Fugu Ultra matches Anthropic’s top model on several benchmarks.
- Anthropic adds Claude Tag for Slack — On June 24 Anthropic shipped a beta that lets Team and Enterprise users tag @Claude into Slack channels to delegate tasks and connect tools, data, and codebases.
- Gemini image preview models reach shutdown — June 25 was the deprecation deadline for the gemini-3.1-flash-image-preview and gemini-3-pro-image-preview endpoints, pushing developers to the GA image models.