Anthropic Launches Opus 4.5 AI, Its Most Advanced AI Model Yet

State-of-the-Art Performance Across Key Benchmarks

Anthropic has unveiled Opus 4.5, the final iteration of its 4.5 AI model series, following earlier releases of Sonnet 4.5 and Haiku 4.5. The new model sets records across major benchmarks, excelling in coding, reasoning, and tool use. It outperforms competitors on SWE-Bench, Terminal-bench, tau2-bench, MCP Atlas, ARC-AGI 2, and GPQA Diamond, becoming the first AI to score above 80% on SWE-Bench—a key measure of coding proficiency. This breakthrough suggests Opus 4.5 can rival or even surpass human software engineers in coding tasks.

Enhanced Computer and Spreadsheet Functionality

To improve productivity workflows, Anthropic has upgraded Claude for Chrome and Claude for Excel, expanding access to these tools. The Chrome extension is now available to all Claude Max users, while the Excel-integrated model will roll out to Max, Team, and Enterprise tiers. These enhancements aim to streamline data analysis and automate repetitive tasks, making AI assistance more seamless in professional environments.

Memory Upgrades Enable “Endless Chat” Feature

Opus 4.5 introduces significant improvements in long-context memory management, addressing a key limitation in AI conversations. According to Dianne Na Penn, Anthropic’s head of product management for research, simply increasing the context window wasn’t enough—the model needed smarter retention of relevant information. As a result, paying users will now benefit from an “endless chat” feature, where older parts of conversations are automatically compressed without disrupting the flow.

Optimized for Agentic Workflows and Enterprise Use

The latest model is particularly strong in agent-based applications, where Opus 4.5 acts as a central coordinator for multiple AI sub-agents (such as Haiku-powered assistants). This makes it ideal for navigating large codebases, reviewing lengthy documents, and handling complex, multi-step tasks. The release comes amid fierce competition, with OpenAI’s GPT-5.1 and Google’s Gemini 3 also vying for dominance in the AI space.