Anthropic rolls out Claude Sonnet 4.6 with advanced coding and reasoning skills

Anthropic is moving at warp speed in the cutthroat AI space. Just weeks after unveiling Claude Opus 4.6, the AI company has now launched Claude Sonnet 4.6 and is positioning it as its most capable Sonnet model yet. The focus this time is sharper coding and reasoning skills, and the ability to handle serious amounts of information without losing the thread.

Claude Sonnet 4.6 is now the default model inside Anthropic’s Claude chatbot, available to both free users and Pro subscribers. Free users, however, get capped access that refreshes every five hours. Developers can also tap into the model via API and through major cloud platforms, so this release is aimed as much at enterprises as everyday chatbot users.

Built for code and complex work

Anthropic describes Sonnet 4.6 as a “full upgrade” across coding, long-context reasoning, agent planning and what it calls knowledge work. In practice, that means tasks such as analysing documents, summarising reports, navigating spreadsheets and assisting with design workflows.

The company also claims the model is more reliable when generating and debugging code, and better at sticking to instructions without wandering off-script, a common frustration with earlier AI systems.

One of the more impressive features is the 1 million token context window, that’s currently in beta. That expanded memory allows the model to process and reference far larger documents or codebases in a single session.

For legal teams combing through contracts, financial analysts parsing long reports, or developers managing sprawling repositories, that kind of context depth could make quite the difference.

Benchmarks and lower hallucination rates

Anthropic says Sonnet 4.6 performed strongly on benchmarks including Humanity’s Last Exam, GPQA Diamond and SWE-bench Verified, which test reasoning depth and coding accuracy. It also points to improved results in insurance and enterprise automation scenarios compared to earlier Claude models.

ALSO READ: Anthropic launches Claude Cowork, an assistant that can clean up your desktop

On the safety front, the company claims lower hallucination rates and reduced “sycophancy”, the tendency for AI systems to agree with user assumptions even when they are wrong.

With Sonnet 4.6 becoming the default experience inside Claude, Anthropic is more than ready for mainstream use. Will you use it for your coding and reasoning tasks? Drop a comment to let us know.