Anthropic launches Claude Opus 4.5

About two months after the introduction of the fast AI model Claude Sonnet 4.5, Anthropic is releasing its new AI model, Claude Opus 4.5. According to Anthropic, it is "the world's most powerful model for programming, autonomous agents, and computer control." It is also said to perform better in daily tasks such as spreadsheet processing, deep research, and presentations. The company states that it outperforms Google's new AI model Gemini 3 in various programming categories.

According to Anthropic, Claude Opus 4.5 is available starting today through Anthropic's apps, the API, and all three major cloud providers.

Opus 4.5 usable in Excel

As Anthropic writes in the blog post announcing the launch, Opus 4.5 is said to be better than its predecessor in research, working with presentations, and filling out spreadsheets. Furthermore, the company has released new tools within Claude Code, its programming tool, and the Claude apps for users that are intended to enable, according to Anthropic, "longer-running agents and new ways to use Claude in Excel, Chrome, and on the desktop."

The paid Claude Max plan is required for the Claude for Chrome feature. With Claude for Excel, a chat is integrated into Microsoft's spreadsheet program. This allows information about spreadsheet content to be queried and edited directly. The latter function can be used by Max, Team, and Enterprise users.

Said to handle ambiguity better

The new AI model is said to handle ambiguity better and be able to weigh trade-offs without assistance. According to Anthropic, this has been confirmed in both internal tests and by customers with early access. Testers found that Claude Opus 4.5 "finds solutions for complex, multi-system errors." Tasks that were nearly impossible for Sonnet 4.5 just a few weeks ago are now within reach.

With Claude Opus 4.5, Anthropic is also introducing a new variable, the "effort parameter," for the programming interface (API). This allows developers to control how intensely the model should think about a task.

With Claude Opus 4.5, Anthropic also addresses security issues with AI agents, such as malicious use cases and prompt injection attacks. In the latter, harmful text is hidden in a website or other data source from which the LLM retrieves data, instructing it to bypass its security measures and, for example, disclose personal data. According to Anthropic, the new model is "harder to trick with prompt injection than any other groundbreaking model in the industry." However, it is still not completely immune.

For Claude and Claude Code users with access to Opus 4.5, the manufacturer has lifted the Opus-specific upper limits. Anthropic has also increased the general usage limits for Max and Team premium users. This is intended to give them approximately the same number of Opus tokens as they had with Sonnet previously. The company has updated the usage limits to ensure that users can use Opus 4.5 for daily work.

(afl)