Anthropic has announced Opus 4.5, the newest version of its flagship model and the final model in its 4.5 series. The release follows Sonnet 4.5 in September and Haiku 4.5 in October, giving Anthropic a full lineup aimed at different levels of AI work.
The company is positioning Opus 4.5 around stronger benchmark results, better tool use, spreadsheet work, computer use, and memory improvements. It is also using the launch to expand Claude for Chrome and Claude for Excel beyond their earlier pilot phase.
A flagship model with benchmark gains
Opus 4.5 arrives with state-of-the-art performance across a range of benchmarks named by Anthropic. Those include coding benchmarks such as SWE-Bench and Terminal-bench, tool-use tests such as tau2-bench and MCP Atlas, and general problem-solving benchmarks including ARC-AGI 2 and GPQA Diamond.
The standout claim is in coding. Opus 4.5 is the first model to score over 80% on SWE-Bench verified, a respected coding benchmark. For a model marketed as a flagship system, that result matters because coding remains one of the clearest ways to test whether an AI model can reason through structured, practical tasks.
Benchmarks do not describe every real-world use case. Still, the mix Anthropic highlighted shows where it wants Opus 4.5 to be judged: software work, tool operation, and complex problem solving. Those are also the areas where the company is pairing the model with new product availability.
Claude moves further into Chrome and Excel
Alongside Opus 4.5, Anthropic is making Claude for Chrome and Claude for Excel more broadly available. Both products were previously in pilot, and both are meant to show how the model performs in everyday work settings rather than only inside a chat window.
The availability differs by product. Claude for Chrome will be available to all Max users. The Excel-focused model will be available to Max, Team, and Enterprise users.
That split reflects two different kinds of use. Chrome puts Claude closer to browser-based tasks and computer use. Excel puts the model in a spreadsheet environment, where Anthropic is emphasizing Opus 4.5's spreadsheet capabilities.
These integrations matter because they place the model inside tools where users already do structured work. Instead of treating the AI system as a separate destination, Anthropic is making Claude available closer to the browser and spreadsheet workflows mentioned in the launch.
Memory becomes a central feature
Anthropic also emphasized memory improvements for long-context operations. According to the source article, those improvements required significant changes in how the model manages memory.
Dianne Na Penn, Anthropic's head of product management for research, told TechCrunch that Opus 4.5 includes improvements to general long context quality in training. But she also said that "context windows are not going to be sufficient by themselves."
The point is that longer context alone does not solve every problem. A model also needs to decide which details matter enough to retain and use later. Penn described knowing the right details to remember as important alongside a longer context window.
Those memory changes support a new feature for paid Claude users: endless chat. When a chat reaches the model's context window, the conversation can continue without interruption. Instead of stopping the user, the model compresses its context memory without alerting the user.
Built for agentic work
Many of the Opus 4.5 upgrades are aimed at agentic use cases. The source article describes scenarios where Opus acts as a lead agent directing a group of Haiku-powered sub-agents.
That setup makes memory more important. A lead agent has to keep track of the larger task, coordinate smaller pieces of work, and know when earlier information should be revisited. Penn said this is where "fundamentals like memory become really important."
She also said Claude needs to explore code bases and large documents, while knowing when to backtrack and recheck something. That description fits the broader direction of the release: Opus 4.5 is not only about answering prompts, but about handling longer, tool-heavy workflows.
- Coding: Opus 4.5 is highlighted for SWE-Bench and Terminal-bench performance.
- Tool use: Anthropic points to tau2-bench and MCP Atlas.
- Long context: The model includes memory changes for long-context operations.
- Products: Claude for Chrome and Claude for Excel are becoming more broadly available.
A crowded frontier model market
Opus 4.5 is launching into a competitive moment for frontier models. The source article notes competition from OpenAI's GPT 5.1, released on November 12, and Google's Gemini 3, released November 18.
That context makes Anthropic's product choices significant. The company is not only presenting benchmark gains. It is also expanding where Claude can be used, adding broader access to Chrome and Excel integrations, and changing how paid Claude chats continue when context limits are reached.
The result is a release that ties model performance to practical work surfaces. Opus 4.5 is being framed as a stronger coding and tool-use model, but also as a system designed for longer sessions, spreadsheet tasks, browser use, and agentic workflows.