The company said on Monday that the update will let Copilot’s Researcher agent draw responses from both OpenAI’s GPT and Anthropic’s Claude models within a single workflow. Until now, it relied on one model for each task.

With the new “Critique” feature, one model handles the initial draft while another reviews and refines it before the final output is produced. 

Microsoft said this setup is designed to improve accuracy and reduce errors that can appear in AI-generated responses.

Microsoft is also adding a feature called “Model Council”, which lets users compare answers from different AI systems side by side. This gives a clearer view of where responses match or differ, and helps users decide which output to rely on.

The company says the changes are aimed at improving speed, quality and overall productivity for users working on complex tasks.

At the same time, Microsoft is expanding access to its Copilot Cowork tool, which is still in early release under its Frontier programme. The tool is built for longer, multi-step tasks. Users can describe what they want done, and the system creates a plan, works through connected steps and shows progress along the way.

Copilot Cowork can also carry out routine work such as scheduling, preparing briefings and handling repeat tasks like monthly reviews. Microsoft said early users are already applying it to planning, document creation and preparation for executive meetings.

The company has been focused on strengthening its Copilot platform as competition grows from other AI tools, including Google’s Gemini and Anthropic’s Claude-based systems.

The company confirmed that Opus 4.5 provides leading results in software engineering tests, becoming the first model to clear the 80% mark on the SWE-Bench Verified assessment. 

It also recorded success in problem-solving tasks and computer-based performance, areas Anthropic has pushed aggressively in recent months.

Alongside the launch, the company is enhancing access to several products built around the model’s capabilities. Claude for Chrome will now be open to all Max users, while Claude for Excel is expanding to Max, Team and Enterprise customers. These tools were previously restricted to pilot testing.

Opus 4.5 introduces a redesigned memory system intended to improve long-running tasks and reduce disruptions during extended conversations. Users with paid plans will now be able to continue chats without being cut off when the system reaches its context limit, as the model compresses earlier exchanges quietly in the background.

Dianne Na Penn, Anthropic’s head of product management for research, noted the change in how the system handles information, saying: “There are improvements we made on general long context quality in training with Opus 4.5, but context windows are not going to be sufficient by themselves. Knowing the right details to remember is really important in complement to just having a longer context window.”

Anthropic has also framed the update as preparation for more advanced agent-style tasks, where Opus directs smaller models to complete complex work across documents and codebases. 

Penn noted the need for stability in those scenarios, adding: “This is where fundamentals like memory become really important, because Claude needs to be able to explore code bases and large documents, and also know when to backtrack and recheck something.”

The release arrives during a crowded period in the sector, coming less than two weeks after OpenAI launched GPT 5.1 and shortly after Google rolled out Gemini 3 on November 18. All three companies are competing for enterprise customers seeking systems capable of specialised technical work.

Opus 4.5 is available through Anthropic’s apps, API and major cloud platforms. The company has also increased usage allowances for Max and Team Premium customers to support daily adoption.

Anthropic says more details on performance and safety testing are outlined in its system documentation, which states that Opus 4.5 is the most secure model the firm has produced so far.