Anthropic has unveiled its new Claude Opus 4.5 model. What can it do?

Anthropic has unveiled its new flagship AI model, Claude Opus 4.5. The company calls it the most powerful model to date: it has significantly improved in programming, working with agents, using computers, as well as in research, analytics, and document processing tasks.

This is stated on the developers’ official website.

Opus 4.5 is already available in Claude applications, via API, and on all major cloud platforms. For developers, the model is called “claude-opus-4-5-20251101” and costs $5 per 1 million input tokens and $25 per 1 million output tokens — lower than previous generations.

According to Anthropic’s test results, Claude Opus 4.5 demonstrated a level unattainable for other models in practical programming scenarios:

• on SWE-bench Verified, the model showed the best performance on the market;

• the model significantly improved accuracy in mathematics, logic, and analytical abilities.

The developers note that the model works better with ambiguous queries, independently finds non-standard solutions, and effectively eliminates complex bugs in the code.

During internal testing, Opus 4.5 even outperformed real candidates taking technical exams for engineering positions at the company.

Another update is enhanced security. According to Anthropic, Opus 4.5 has become the most resistant of the company’s models to attacks directly in prompts and other forms of manipulation.

Along with the new model, the company has updated the Claude Developer Platform, the Claude Code service, and features in the browser extension and desktop application.

In particular, an effort parameter has been added, allowing developers to choose the balance between speed, cost, and maximum quality of task execution.

Read more: Anthropic raised $15B in investments from Microsoft and NVIDIA and buys Azure for $30B