Anthropic has announced the launch of two new artificial intelligences – Claude Opus 4 and Claude Sonnet 4. They are focused on performing complex tasks, writing code and integrating with internet search. According to the developers, the flagship model Opus 4 has become the most powerful in the Claude series and is able to work autonomously for up to seven hours.
Opus 4 and Sonnet 4 show significant progress compared to previous generations – 65 per cent less likely to look for “workarounds” to tasks and better at operating information during long sessions, especially when accessing the user’s local files.
Anthropic’s internal tests showed the Opus 4 outperforming competitors including Google Gemini 2.5 Pro, OpenAI o3 reasoning and GPT-4.1 in tasks involving programming and using tools such as web search.
The Sonnet 4, which replaced version 3.7, is optimised for more general tasks. It has become more accurate, improving answer logic and code writing skills. Among the new features-thinking summaries that summarise the AI’s thought process in a convenient way-and a beta version of extended thinking, which allows the model to switch between analytical mode and tool search mode to improve accuracy.
Both models-Opus 4 and Sonnet 4-are already available through the ANTHROPIC API, Amazon Bedrock and Google Cloud (Vertex AI). They are included in Claude’s paid subscription, but Sonnet 4 is also available for free. Advanced features, including extended thinking, remain exclusive to subscribers.
The company has also opened up general access to Claude Code, a team interface for developers that was previously in limited testing. Following a February pre-release, the tool is now available to all users.
Anthropic has also signalled its intention to update its models more frequently to keep competing with players such as OpenAI, Google and Meta.