On Monday, Anthropic announced Opus 4.5, the latest version of its flagship model. This is the final model in Anthropic’s 4.5 series to be released, following the launch of Sonnet 4.5 in September and Haiku 4.5 in October.
As expected, the new Opus version demonstrates state-of-the-art performance across a range of benchmarks. These include coding evaluations such as SWE-Bench and Terminal-bench, tool use assessments like tau2-bench and MCP Atlas, and general problem solving tests including ARC-AGI 2 and GPQA Diamond. A notable achievement is that Opus 4.5 is the first model to score over eighty percent on SWE-Bench verified, a respected coding benchmark.
Anthropic also highlighted the model’s capabilities for computer use and working with spreadsheets. To showcase these functions, the company launched several parallel products. Alongside Opus 4.5, Anthropic will make its Claude for Chrome and Claude for Excel products more broadly available; these were previously in pilot programs. The Chrome extension will be available to all Max users, while the Excel-focused model will be available to Max, Team, and Enterprise users.
Opus 4.5 includes memory improvements for long-context operations, which required significant changes to how the model manages its memory. Dianne Na Penn, Anthropic’s head of product management for research, explained that while there are general long context quality improvements from training, context windows alone are not sufficient. She emphasized that knowing the right details to remember is critically important in addition to having a longer context window.
These memory changes also enabled a long-requested endless chat feature for paid Claude users. This feature allows chats to continue without interruption when the model reaches its context window. Instead of stopping, the model will compress its context memory without alerting the user.
Many of the upgrades are designed with agentic use-cases in mind, particularly scenarios where Opus acts as a lead agent commanding a group of Haiku-powered sub-agents. Managing these complex tasks requires a strong command of working memory, which is where the memory improvements prove their value. Penn noted that fundamentals like memory are essential because Claude needs to explore code bases and large documents, and also know when to backtrack and recheck something.
Opus 4.5 will face stiff competition from other recently released frontier models, most notably OpenAI’s GPT 5.1 released on November 12 and Google’s Gemini 3 released on November 18.