Anthropic released Claude Sonnet 5 on June 30, 2026, a midtier model designed to bring near-flagship agentic capabilities to developers at lower cost (TechCrunch). At launch the model is priced at $2 per million input tokens and $10 per million output tokens on an introductory basis through August 31; standard pricing afterward moves to $3 and $15 per million tokens respectively (MacRumors, Anthropic). On a knowledge-work benchmark Sonnet 5 slightly outperforms Opus 4.8, while benchmarks for reasoning, tool use, and software coding show substantial improvement over its predecessor Sonnet 4.6 (9to5Mac). The model ships as the default for Free and Pro users, with access extended to Max, Team, Enterprise, Claude Code, and the API from day one (Anthropic). Anthropic also reports improved resistance to prompt-injection attacks and lower rates of compliance with misuse requests relative to the prior version - properties directly relevant to teams running autonomous agent pipelines (MacRumors).
Also on June 30, the U.S. Department of Commerce lifted export controls it had imposed on Anthropic’s Fable 5 and Mythos 5 models, clearing the way for global access to resume (CNBC, The Washington Post). The controls had been imposed around June 13, requiring Anthropic to immediately suspend access for all foreign nationals - including its own employees - after the Commerce Department cited a jailbreak it said could enable the models to act as unrestricted cyber tools (Fortune). Anthropic had publicly contested that characterization, describing the jailbreak as specific to a single narrow scenario rather than a broad bypass of Fable 5’s safety guardrails (Forbes). Both models are scheduled to return to Claude.ai, Claude Code, and the API starting July 2.