On Monday, Anthropic introduced Opus 4.5, the newest model of its flagship mannequin. It’s the final of Anthropic’s 4.5 sequence of fashions to be launched, following the launch of Sonnet 4.5 in September and Haiku 4.5 in October.
As anticipated, the brand new model of Opus has state-of-the-art efficiency on a spread of benchmarks, together with coding benchmarks (SWE-Bench and Terminal-bench), device use (tau2-bench and MCP Atlas), and basic drawback fixing (ARC-AGI 2, GPQA Diamond).
Notably, Opus 4.5 is the primary mannequin to attain over 80% on SWE-Bench verified, a revered coding benchmark.
Anthropic additionally emphasised Opus’ laptop use and spreadsheet capabilities, and launched numerous parallel merchandise to showcase how the mannequin holds up in these settings. Along with Opus 4.5, Anthropic will make its Claude for Chrome and Claude for Excel merchandise — beforehand in pilot — extra broadly accessible. The Chrome extension will probably be accessible to all Max customers, whereas the Excel-focused mannequin will probably be accessible to Max, Crew, and Enterprise customers.
Opus 4.5 additionally comes with reminiscence enhancements for long-context operations, which required vital adjustments in how the mannequin manages its reminiscence.
“There are enhancements we made on basic lengthy context high quality in coaching with Opus 4.5, however context home windows are usually not going to be adequate by themselves,” Dianne Na Penn, Anthropic’s head of product administration for analysis, instructed Trendster. “Figuring out the suitable particulars to recollect is admittedly vital in complement to simply having an extended context window.”
These adjustments additionally enabled a long-requested “countless chat” characteristic for paid Claude customers, which can enable chats to proceed with out interruption when the mannequin hits its context window. As an alternative, the mannequin will compress its context reminiscence with out alerting the consumer.
Techcrunch occasion
San Francisco
|
October 13-15, 2026
Lots of the upgrades are made with a watch towards agentic use instances, significantly eventualities wherein Opus acts as a lead agent commanding a bunch of Haiku-powered sub-agents. Managing these duties requires a powerful command of working reminiscence, which is the place the reminiscence enhancements described by Penn actually present their price.
“That is the place fundamentals like reminiscence turn out to be actually vital,” Penn says, “as a result of Claude wants to have the ability to discover code bases and huge paperwork, and in addition know when to backtrack and recheck one thing.”
Opus 4.5 will face stiff competitors from different just lately launched frontier fashions, most notably OpenAI’s GPT 5.1 (launched on November 12) and Google’s Gemini 3 (launched November 18).
