Anthropic PBC unveiled Claude Sonnet 5, a midrange large language model that advances coding abilities and safety measures over its predecessor, and reinstated access to its previously restricted Fable 5 and Mythos 5 models following the removal of US export controls.
- Sonnet 5 improves coding scores by up to 13.4% and autonomously double-checks output.
- Built-in guardrails reduce cybersecurity risks, including prompt injection attacks.
- US export controls lifted, allowing renewed rollout of Mythos 5 and wider release of Fable 5.
What happened
Anthropic introduced Claude Sonnet 5, a mid-tier large language model designed to replace its predecessor in the company’s consumer chatbot service tiers. This model demonstrated significant improvements in coding performance as measured by SWE-Bench Pro and Terminal-Bench 2.1, surpassing the previous Sonnet version by 5.1% and 13.4%, respectively. Sonnet 5 also features enhanced autonomy, including self-verification of outputs without user prompting.
In addition to the Sonnet 5 release, Anthropic announced that US government export controls on its advanced language models Mythos 5 and Fable 5 have been lifted. This regulatory change allows the company to resume offering these models, with Fable 5 planned for broad availability and Mythos 5 restricted to select trusted organizations.
Why it matters
The improved coding capabilities and autonomous behaviors of Sonnet 5 position it as a competitive midrange model in the increasingly crowded LLM market. While OpenAI’s competing Terra model outperforms Sonnet 5 slightly on one benchmark, Anthropic’s focus on safety enhancements differentiates its offering. Sonnet 5 includes stronger defenses against cybersecurity threats such as prompt injection attacks, mitigating risks posed by increased model autonomy.
Lifting export controls on Mythos 5 and Fable 5 is a significant development amid ongoing regulatory scrutiny of advanced AI technologies. This enables Anthropic to expand access to their more powerful and safety-guarded models, potentially accelerating adoption across enterprise and developer segments while maintaining legal compliance.
What to watch next
Anthropic plans to make Claude Sonnet 5 available across its consumer tiers including free, Pro, Max, Team, and Enterprise plans starting immediately, with developer API access priced at $3 per million input tokens and $15 per million output tokens from September. Market observers will be tracking user adoption and feedback, especially relating to Sonnet 5’s balance of performance and safety.
The gradual rollout of Mythos 5 and Fable 5 will be closely watched as these models represent Anthropic’s flagship and guarded AI solutions, respectively. Evaluating their real-world use cases and compliance with new regulatory frameworks will be critical in understanding how export control changes influence competitive dynamics in large language models.