Anthropic has introduced Claude Opus 4.5, an AI model positioned as its most advanced system for coding and enterprise agent applications.

The release follows recent updates to the company’s Haiku and Sonnet series and is designed for users seeking AI support on complex office, coding, and workflow activities.

According to Anthropic, Claude Opus 4.5 model is intended to be used as a partner for professional tasks rather than a simple content generator.

The new model incorporates features aimed at improving productivity in document, spreadsheet, and presentation creation.

Anthropic noted that Claude Opus 4.5 can perform repetitive office tasks by interacting directly with computers and browsers.

The company also highlighted new tools for developing long-running agents, expanded integration options for Excel and Chrome, and improved capacity for extended conversations without context window limit errors.

Claude Opus 4.5 is now accessible through a range of platforms. The Claude for Chrome extension allows the AI to take actions across browser tabs and is available to all Max users.

For spreadsheet users, Claude for Excel introduces sidebar chat support within the application. This feature is now generally available to Max, Team, and Enterprise users and includes enhancements such as pivot table, chart creation, and file upload support.

The Infinite Chat feature enables the model to maintain context over longer interactions by using memory across files, which addresses frequent user complaints about context limitations on paid plans.

Developers can access new updates, including Claude Code on desktop in a research preview phase and additional developer platform capabilities.

Anthropic reported that the full Claude 4.5 generation supports various stages of development.

Opus 4.5 is geared towards production code and lead agent use cases, Sonnet 4.5 supports rapid prototyping and scaled deployments, while Haiku 4.5 targets sub-agents and the company’s free product tier.

The company identified various use cases where Claude Opus 4.5 advances performance compared to earlier models.

In autonomous agents, Anthropic claims that Opus 4.5 requires only four iterations to reach peak capability while other systems struggle after ten rounds. These agents also retain insights from prior tasks for future application.

Rakuten business AI general manager Yusuke Kaji said: “Claude Opus 4.5 represents a breakthrough in self-improving AI agents. For automation of office tasks, our agents were able to autonomously refine their own capabilities—achieving peak performance in 4 iterations while other models couldn’t match that quality after 10.

“They also demonstrated the ability to learn from experience across technical tasks, storing insights from past work and applying them to new challenges like SRE operations.”

In software engineering contexts, the model is described as handling ambiguous coding issues without detailed user supervision and is able to identify solutions that involve reasoning across multiple systems.

For enterprise environments combining information retrieval, tool usage, and analysis, Anthropic cites state-of-the-art benchmark results and mentions AI workspace firm Genspark as having measured improved performance on multi-step reasoning tasks with Claude Opus 4.5.

In financial modelling use cases within Excel, early adopters reportedly observed a 20% improvement in accuracy on internal tests, along with a 15% gain in efficiency, and the ability to complete more complex assignments.

Genspark chief technology officer Kay Zhu said: “Claude Opus 4.5 achieved state-of-the-art results for complex enterprise tasks on our benchmarks, outperforming previous models on multi-step reasoning tasks that combine information retrieval, tool use, and deep analysis.”

Recently, Anthropic committed to purchasing $30bn worth of compute capacity from Microsoft Azure, as part of newly announced strategic partnerships with both Microsoft and Nvidia.