Between 30 June and 1 July 2026, Anthropic had a remarkably busy week. On 30 June the company launched Claude Sonnet 5, its new mid-tier model. On the same day it debuted Claude Science, an AI workbench for researchers. Then on 1 July, Claude Fable 5, the model that had been effectively unavailable for weeks, officially came back online.
Fable 5 was sidelined after a government directive tied to national security concerns. Its return follows a long negotiation with the US government, which now allows a select group of approved customers to use it under tight controls. Anthropic has been careful to frame the return as a compromise: the model is back, but not freely available, and safety rules around it remain strict.
Claude Sonnet 5 is the practical release for a wider audience. Anthropic says it can plan, use tools like browsers and terminals, and run autonomously at a level that just a few months ago required larger and more expensive models. The company also notes that it has much lower ability to perform dangerous cybersecurity tasks than its bigger Opus siblings.
Claude Science, meanwhile, is not a new model but a new product. It pulls fragmented tools and datasets into one place for scientists, and can generate figures, charts, and even 3D protein structures. Anthropic says it is starting with biology and will expand from there.
Taken together, the three moves show where Anthropic is heading: more autonomous agents, more domain-specific tools, and closer engagement with government. The company is no longer just selling a chatbot. It is building a full platform for AI work in science, business, and government, while trying to keep its safety reputation intact.