Anthropic has announced Opus 4.5, the latest and final model in its 4.5 series, arriving after the earlier releases of Sonnet 4.5 in September and Haiku 4.5 in October. The new flagship model delivers state-of-the-art performance across a wide range of industry benchmarks, including coding, tool use, and advanced problem-solving.
Opus 4.5 became the first model ever to score above 80% on SWE-Bench verified, a respected coding benchmark, underscoring its improved reasoning and coding capabilities. It also achieved top results on tests like SWE-Bench, Terminal-bench, tau2-bench, MCP Atlas, ARC-AGI 2, and GPQA Diamond.
Beyond benchmarks, Anthropic highlighted Opus 4.5’s strengthened skills in computer use and spreadsheet operations, unveiling two new companion products: Claude for Chrome and Claude for Excel. Both tools are moving out of pilot, with the Chrome extension rolling out to Max users and the Excel version expanding to Max, Team, and Enterprise plans.
One of the biggest advancements in Opus 4.5 is its upgraded memory system for long-context tasks. According to Dianne Na Penn, Anthropic’s head of product management for research, context length alone is not enough, models must also decide which details truly matter. This redesign enables smoother long-form interactions and reduces context drop-offs.
These memory updates also power a highly anticipated feature: “endless chat.” Paid Claude users can now continue conversations without interruptions when the model reaches its context window. Instead, Opus 4.5 will intelligently compress past conversation data without notifying the user.
Related: Anthropic Expands Google Cloud Partnership to Power Next-Gen AI with One Million TPUs
Anthropic built many of these improvements with agentic workflows in mind. Opus 4.5 is designed to function as a lead agent capable of directing smaller Haiku-based sub-agents, a setup that requires strong working memory, the ability to explore large documents and codebases, and reliable backtracking.
With its release, Opus 4.5 enters a competitive landscape dominated by other frontier AI models, including OpenAI’s GPT 5.1, released November 12, and Google’s Gemini 3, launched November 18.
