Anthropic PBC today launched Claude Opus 4.5, its new flagship large language model.

The company says Opus 4.5 is its safest and most capable LLM yet. The model is rolling out a few weeks after the two other entries into the Claude 4.5 series: Sonnet 4.5 and Haiku 4.5. The LLMs are positioned as midrange and entry-level alternatives to Opus 4.5, respectively. 

According to Anthropic, Opus 4.5 is better than the competition at powering artificial intelligence agents that use tools to automate work. When agents based on the model encounter a task they can’t complete on their first try, they can iteratively refine their capabilities. Anthropic says that Opus 4.5 reaches “peak performance” after four iterations, while other LLMs require 10 attempts.

Opus 4.5 also brings other improvements. Compared to Anthropic’s other models, it provides better support for long-running agents. That should make the LLM more useful for tasks such as rewriting applications that can take several hours.

Developers often automate complex, long-running tasks using not one but several agents that coordinate their work. According to Anthropic, software teams pursuing such projects can use Opus 4.5 to power the lead agent and the entry-level Haiku 4.5 to power sub-agents. Assigning simple processing steps to a lightweight LLM lowers inference costs.

Programming is another area where Optus 4.5 provides better performance than its predecessors. According to Anthropic, the model requires less human guidance and handles ambiguity better. A developer could, for example, ask Opus 4.5 to troubleshoot a bug without specifying that fixing it requires the model to review multiple systems.

The model’s new reasoning features are complemented by integrations with Excel and Google Chrome.

Anthropic first introduced the Excel integration last month as a research preview. The add-on rolled out to Claude for Financial Services, a feature bundle geared towards financial professionals. It makes Claude accessible through a sidebar in the Excel interface.

The integration is now generally available for users with Max, Team, and Enterprise subscriptions. Anthropic has added support for pivot tables, tables that are used to summarize information spread across a large number of spreadsheet fields. Additionally, users can now upload files and generate charts.

Claude’s Chrome integration debuted two months before the Excel add-on. It’s a browser extension that enables the chatbot to perform actions in web applications on the user’s behalf. At the time of its introduction, Anthropic stated that the feature includes mitigations against malicious prompts embedded in web content.

The Chrome extension initially rolled out to 1,000 users of Claude’s top-end Max plan. Today, Anthropic made the feature generally available to all Max subscribers. 

The company also released a number of other enhancements as part of today’s update. The Claude Code programming assistant is now available in Anthropic’s desktop client, while the Max and Teams plans have received higher usage caps. Claude Chat, in turn, is gaining the ability to summarize information from earlier parts of a chat session.

Opus 4.5 is available through Claude Chat, Claude Code and application programming interfaces. Developers who use the APIs have access to a new “effort” setting that makes it possible to adjust the amount of time and computing capacity the LLM uses to perform a task. The more infrastructure is allocated to a task, the higher the output quality.

Opus 4.5 is priced at $5 per million input tokens and $25 per million output tokens.

Image: Anthropic