Anthropic Releases Opus 4.5 With New Chrome And Excel Integrations

On Monday, Anthropic announced Opus 4.5, nan latest type of its flagship model. It’s nan past of Anthropic’s 4.5 bid of models to beryllium released, pursuing nan motorboat of Sonnet 4.5 successful September and Haiku 4.5 successful October.

As expected, nan caller type of Opus has state-of-the-art capacity connected a scope of benchmarks, including coding benchmarks (SWE-Bench and Terminal-bench), instrumentality usage (tau2-bench and MCP Atlas) and wide problem solving (ARC-AGI 2, GPQA Diamond). 

Notably, Opus 4.5 is nan first exemplary to people complete 80 percent connected SWE-Bench verified, a respected coding benchmark.

Anthropic besides emphasized nan Opus’s machine usage and spreadsheet capabilities, and launched a number of parallel products to showcase really nan exemplary holds up successful those settings. Together pinch Opus 4.5, Anthropic will make its Claude for Chrome and Claude for Excel products — antecedently successful aviator — much broadly available. The Chrome hold will beryllium disposable to each Max users, while nan Excel-focused exemplary will beryllium disposable to Max, Team and Enterprise users.

Opus 4.5 besides comes pinch representation improvements for long-context operations, which required important changes successful really nan exemplary manages its memory.

“There are improvements we made connected wide agelong discourse value successful training pinch Opus 4.5, but discourse windows are not going to beryllium capable by themselves,” Dianne Na Penn, Anthropic’s caput of merchandise guidance for research, told TechCrunch. “Knowing nan correct specifications to retrieve is really important successful complement to conscionable having a longer discourse window.”

Those changes besides enabled a long-requested “endless chat” characteristic for paid Claude users, which will let chats to proceed without interruption erstwhile nan exemplary hits its discourse window. Instead, nan exemplary will compress its discourse representation without alerting nan user.

Many of nan upgrades are made pinch an oculus towards agentic use-cases, peculiarly scenarios successful which Opus acts arsenic a lead supplier commanding a group of Haiku-powered sub-agents. Managing those tasks requires a beardown bid of moving memory, which is wherever nan representation improvements described by Penn really show their worth.

“This is wherever fundamentals for illustration representation go really important,” Penn says, “because Claude needs to beryllium capable to research codification bases and ample documents, and besides cognize erstwhile to backtrack and recheck something.”

Opus 4.5 will look stiff title from different precocious released frontier models, astir notably OpenAI’s GPT 5.1 (released connected November 12) and Google’s Gemini 3 (released November 18).