Anthropic releases Sonnet 5, Fable 5 and Mythos 5 to become usable again
Anthropic has not only released a better Sonnet model with a sweep, but also announced that Fable and Mythos are to return.
(Image: Anthropic)
With Claude Sonnet 5, Anthropic has released a supposedly noticeably improved version of its AI model for the most extensive use cases, which is said to catch up to or even surpass the high-end Opus 4.8 model in sub-disciplines. The main improvement has been in its agentic functionality, which means tasks are to be completed independently. The new model “can make plans, use tools like browsers and the console, and operate autonomously to an extent that just months ago required larger and more expensive models,” Anthropic writes. In addition, there is a toolbox for research and the announcement that export restrictions for Fable 5 and Mythos 5 are to be lifted.
Claude Sonnet is Anthropic's AI model that promises the best possible performance for the broadest use cases. For those using Anthropic's AI technology for free or in the Pro offer, the model is the pre-selected standard. According to several benchmarks published by the AI company, there is a sometimes significant improvement compared to the predecessor, with which Sonnet is catching up to Opus 4.8. In agentic tasks, the top model, Opus 4.8, is even almost invariably surpassed. This has been confirmed in external tests; Sonnet now successfully completes tasks that were previously aborted. The model also checks the results without being prompted. All of this is available at better prices.
Fable and Mythos to be re-released
Anthropic also assures that the new model promises more security. Sonnet 5 is better at rejecting malicious requests and fending off takeover attempts. Furthermore, the model hallucinates less often, i.e., produces false outputs, than its direct predecessor. However, in exploiting security vulnerabilities, the model lags far behind Anthropic's most powerful model, Mythos, but is still ahead of Sonnet 4.6. Therefore, Anthropic has also implemented guardrails here to ensure that Sonnet rejects assistance for cyberattacks just like the current Opus models. On Wednesday, Anthropic also released Claude Science, which combines tools for research.
Videos by heise
In parallel with the release of Sonnet 5, Anthropic has also made it public that the US government has lifted the export control restrictions for Mythos and Fable. Therefore, the restrictions should be lifted starting Wednesday and access restored. The AI company thanks users for their patience and announces further information. The move had been foreshadowed over the weekend when a corresponding letter became public. The blocking of Anthropic's most powerful AI models for foreigners had made it clear, especially in Europe, how great the dependence on AI technology from the USA is here.
(mho)
