Anthropic has officially rolled out Opus 4.5, the newest version of its flagship AI model and the final release in the company’s 4.5 series. It follows the earlier launches of Sonnet 4.5 in September and Haiku 4.5 in October, completing the trio that forms Anthropic’s most capable generation of AI systems so far. With Opus 4.5, Anthropic is clearly targeting real-world performance, deeper integrations, and the future of agentic AI.
Opus 4.5 comes with a benchmark performance across multiple categories. It posts standout results in coding benchmarks like SWE-Bench and Terminal-Bench, tool-use tests such as tau2-bench and MCP Atlas, and high-level reasoning sets like ARC-AGI 2 and GPQA Diamond. Most notably, Opus 4.5 becomes the first model ever to surpass 80% on SWE-Bench Verified, a respected coding benchmark often used to measure a model’s ability to fix real-world software issues. This milestone signals a significant leap in the model’s ability to navigate, understand, and resolve complex codebases.
Alongside raw performance, Anthropic is highlighting the model’s improvements in computer use, browser interaction, and spreadsheet intelligence. To showcase these capabilities, the company is expanding access to two supporting tools: Claude for Chrome and Claude for Excel. Both tools were previously in pilot phases, but the launch of Opus 4.5 marks their broader availability.
The Chrome extension is now open to all Claude Max users, allowing Opus 4.5 to interact with webpages, analyze online content, automate browser tasks, and streamline research-heavy workflows. Meanwhile, the Excel-focused model is rolling out to Max, Team, and Enterprise users, bringing enhanced spreadsheet manipulation, formula comprehension, multi-sheet context awareness, and business-ready data handling.
Under the hood, Opus 4.5 includes a major upgrade to long-context memory management. According to Anthropic’s head of product management for research, Dianne Na Penn, simply increasing context window size isn’t enough. Models must also know which details to retain, prioritize, and recall. Significant architectural changes were made to improve what the system remembers and how it compresses older information without disrupting the user experience.
These improvements power the long-requested “endless chat” feature. Instead of cutting off the user mid-conversation when the model reaches its limit, Opus 4.5 quietly compresses earlier messages and keeps the discussion going. This is valuable for users working on multi-stage projects, long documents, large datasets, extensive analyses, or ongoing research tasks where continuity is crucial.
Anthropic also made these upgrades with agentic use-cases in mind. Many organizations are exploring setups where Opus acts as a lead coordinator, overseeing smaller Haiku-powered sub-agents to complete multi-step technical tasks. For such scenarios, strong working memory and the ability to revisit codebases and lengthy documents are essential. Penn notes that memory fundamentals play a key role in enabling the model to know when to backtrack, verify earlier steps, or reassess previous conclusions.
However, Opus 4.5 enters a competitive AI landscape. The release comes shortly after two major frontier model launches: OpenAI’s GPT-5.1, released on November 12, and Google’s Gemini 3, which arrived on November 18. Each model is vying for dominance in agentic workflows, coding reliability, and high-stakes enterprise use.
Still, the combined upgrades in reasoning, browser use, spreadsheet intelligence, memory, and integration tools indicate that Anthropic is positioning Opus 4.5 not just as a model, but as a full productivity engine. With enhanced Chrome and Excel support and a focus on long-term, multi-step work, Opus 4.5 represents Anthropic’s strongest push yet toward AI that actively participates in everyday professional tasks and not just responds to them.
