By Lisa Eadicicco, CNN
(CNN) — Anthropic’s Cowork AI assistant sent shockwaves through Wall Street this week. Now Anthropic is taking another leap forward, improving its model.
Anthropic’s new Claude Opus 4.6 model, announced Thursday, is designed to make Cowork AI better for office and coding work, potentially raising even more concerns that the AI tool could replace specialized software packages that companies use for those tasks.
Legal and financial analysis software stocks have plunged in recent days, bringing the broader stock market down with them. The Nasdaq just had its worst two-day tumble since April, and it’s down another 0.7% Thursday.
Many experts wonder whether AI will ultimately cost some workers their jobs. Tech giants like Anthropic, OpenAI and Google are locked in a race to build AI models that they hope will underpin the future workplace.
Anthropic rival OpenAI just introduced a new platform for creating AI agents meant to function like colleagues Thursday morning. And Anthropic launched its Cowork tool in January.
It remains uncertain whether AI investments will pay off for businesses adopting the technology. Anthropic is betting that its new model will help it replicate the massive success of its Claude Code software but for other types of office work.
“We think that Opus 4.6 is going to be an inflection point for knowledge work in many ways,” Dianne Penn, head of product management for research, said in an interview with CNN ahead of the announcement.
What’s new
One of those ways involves how Opus 4.6 processes data. Anthropic says it’s expanded Opus’ context window, which is the amount of information a model can remember at once, from 200,000 tokens to one million. Tokens are a unit of measurement referring to how AI models understand text. The longer and more complex a query is, the more tokens it requires.
Giving Claude the ability to process more information at once should enable it to handle more complicated tasks, like making sweeping changes to entire code bases, said Penn.
With Opus 4.6, Claude will also be better at knowing when it should take more time to think through a request – a technique known as reasoning – and when it should answer quickly.
Anthropic also says the new model outperforms OpenAI’s GPT-5.2 model on a benchmark evaluating how AI handles knowledge work in fields like finance and legal.
The new PowerPoint integration, available in research preview, will let users build slides using Claude, with the AI able to read layouts and fonts to create slides that align with the desired corporate template. This was particularly challenging because unlike Excel, which is primarily data-driven, PowerPoint slides involve making judgements about design elements like colors and text placement, said Penn.
Anthropic says files that Claude Opus 4.6 works on, like documents, spreadsheets and slides, will be closer to “production-ready” on the first try, meaning they should require less human intervention.
While Anthropic’s Opus 4.6 makes Claude better at non-technical work, the update also includes some improvements for software engineers. Coding duties can be split across teams of agents instead of having one agent work through individual tasks, mimicking the way a human engineering team would operate, Anthropic says.
AI shakes the software industry
The release comes after software stocks cratered this week following the release of plugins for Anthropic’s Cowork tool last Friday. These plugins, which let users tailor Anthropic’s Cowork tool for specific industries like legal, finance