Anthropic unveiled a lower-cost frontier AI model, launched a research-focused AI workbench for scientists and restored global access to its advanced Fable 5 model after U.S. export restrictions were lifted.
Anthropic announced Claude Sonnet 5, positioning it as its new flagship model for coding, AI agents and enterprise knowledge work. The startup said Sonnet 5 delivers performance approaching its larger Opus-class models while lowering costs to $2 per million input tokens and $10 per million output tokens through Aug. 31 and then pricing it at $3 and $15 respectively. Sonnet 5 is available through Claude.ai, the Anthropic API and major cloud platforms.
Anthropic also launched Claude Science, a customizable AI workbench for researchers. Rather than functioning as a general-purpose chatbot, the platform integrates scientific software, programming packages and computing tools into a single environment that produces auditable research artifacts. Anthropic said the product is designed to support research workflows while allowing scientists to customize the environment for their disciplines.
Anthropic announced that Claude Fable 5 has returned to global availability beginning July 1 after the U.S. government lifted export controls imposed in early June over cybersecurity concerns. The startup said it strengthened the model’s safeguards by deploying an improved safety classifier that blocks the reported jailbreak technique in more than 99% of cases. Anthropic also said Mythos 5’s restrictions have been lifted.
Fable 5 and Mythos 5 share the same underlying Mythos-class model architecture. Fable 5 is the broadly available version and incorporates additional safety mechanisms, including classifiers that can block or reroute requests involving potentially harmful cybersecurity tasks.
Mythos 5, which has fewer safeguards, is available only to a limited number of trusted organizations through Anthropic’s Project Glasswing program for defensive cybersecurity work because of concerns its advanced cyber capabilities could be misused. U.S. export restrictions were lifted on June 30.
Meanwhile, Anthropic proposed an industry framework for evaluating AI jailbreak severity, saying it is developing the effort with partners including Amazon, Google and Microsoft to establish common standards for assessing and communicating model safety risks.