Anthropic releases Claude Sonnet 4.6 model, highlighting these improvements

Anthropic rolls out Claude Sonnet 4.6 across plans, highlighting gains in coding, spreadsheet navigation, long-term reasoning, and resistance to prompt injection attacks

Anthropic has announced the launch of Claude Sonnet 4.6. According to the company, the new model introduces upgrades over the Sonnet 4.5 model in aspects like coding, computer use, long-context reasoning, agent planning, knowledge work, and design. The company has also introduced a 1 million token context window in beta, allowing the model to process significantly larger amounts of information in a single session.

 

Anthropic has said that Claude Sonnet 4.6 is now the default model for users on Free and Pro plans via Claude’s website and Claude Cowork. Pricing remains unchanged from Sonnet 4.5, starting at $3/$15 per million tokens.

 

Claude Sonnet 4.6: Availability

Claude Sonnet 4.6 is available across all Claude plans, including the free tier, as well as through Claude Code, Claude Cowork, Anthropic’s API, and select cloud platforms. The free tier has been upgraded to use Sonnet 4.6 by default.

Claude Sonnet 4.6: What’s new

Larger context window

The 1M token context window, currently in beta, is said to allow users to input entire codebases, long legal contracts, or multiple research papers in one go. According to Anthropic, the model is designed not just to accept larger inputs but to reason effectively across them, which could improve long-term planning and complex task execution.

Improved coding performance

Anthropic claims Sonnet 4.6 brings major gains in coding ability. In internal testing within Claude Code, users reportedly preferred Sonnet 4.6 over Sonnet 4.5 around 70 per cent of the time. 

Users also reportedly preferred it over Claude Opus 4.5, citing better instruction following, fewer hallucinations, and more consistent execution of multi-step tasks. 

Computer use capabilities expand

Anthropic has enhanced Claude’s ability to operate a computer interface — claiming that the AI can interact with software the way a person would, by clicking and typing rather than relying on APIs. 

Anthropic says early users of Sonnet 4.6 have observed near-human-level capability in tasks such as navigating complex spreadsheets and completing multi-step web forms across browser tabs. However, the company notes that the model still lags behind highly skilled human users. 

Anthropic also acknowledged security concerns, such as prompt injection attacks, where hidden instructions on websites attempt to manipulate AI behaviour. The company said Sonnet 4.6 shows improved resistance compared to Sonnet 4.5.

Agentic AI and long-horizon reasoning

Anthropic says Sonnet 4.6 demonstrates stronger long-term planning capabilities. 

In a simulated business evaluation called Vending-Bench Arena, the model reportedly adopted a strategy of investing heavily early on before pivoting to profitability later — a move Anthropic says helped it outperform competing models in that test.

 

  The company maintains that its Opus 4.6 model remains better suited for tasks requiring the deepest reasoning, such as complex codebase refactoring or coordinating multiple AI agents.