Anthropic has released “Claude Sonnet 4.6,” an updated version of its core AI model for agent programming and management. The company says Sonnet’s performance is approaching that of its flagship “Opus.”
When working with code, Sonnet 4.6 handles context better, duplicates logic less, and makes code changes more carefully. In Anthropic’s testing, users chose 4.6 over 4.5 for coding-related tasks ~70% of the time. In some tasks, the Sonnet 4.6 also outperforms the Opus 4.5, the previous version of the company’s flagship model.
Another important aspect is that the context window in Sonnet 4.6 has grown to 1 million tokens, although it is currently only in beta. This size allows an AI model to work with entire repositories or dozens of scientific articles.
In addition, the company notes the security of the Sonnet 4.6: in a test of resistance to rapid injection (hidden instructions that can take over a model), the new model performed almost as well as the Opus 4.6.
Sonnet 4.6 is the default model available to all claude․ai and Claude Cowork users, both free and paid versions.
API prices have not increased:
- from $3 for 1 million input tokens
- from $15 for 1 million output tokens
Credits:
All images courtesy of Anthropic.
