Why Anthropic just doubled Claude Cowork limits at no charge

Anthropic has announced a limited-time promotion that doubles users’ five-hour usage limits in Claude Cowork.

Announced on Monday, it’s sure to win over some users with a promise of a double-stuffed helping at no extra charge during this period of tokenomics malaise. Claude Cowork is the company’s collaborative development environment, built for non-technical knowledge workers, and it can synthesize information from multiple sources to complete tasks without requiring the user to coordinate each step.

Double for free — do you mean for me?

Also used by software developers to shoulder tasks related to documents, data, and files, this offer is available to users on Pro, Max, and Team plans. It also extends to legacy seat-based users on Enterprise plans. 

Free plans and consumption-based Enterprise seats are not included in this promotion.

The exact terms stated by Anthropic inform us that, “From June 5, 2026, through July 5, 2026, your 5-hour usage limit in Cowork is doubled. Weekly usage limits remain unchanged. No action is required to participate. If you’re on an eligible plan, the doubled usage is automatically applied.”

This is Claude Cowork, not Claude

The 2x usage increase applies only to Claude Cowork. Usage limits for other Claude products — such as Claude on web, desktop, and mobile — remain unchanged.

Launched on January 12 of this year as a research preview, Claude Cowork allows users to point the tool at a specific folder, set of local files, or application(s) to return a finished deliverable. Users can modify the software’s subsequent actions through a standard chat interface.

“These kinds of promotions are a classic top-of-funnel move: Get developers and users at large hooked on the workflow before they’ve had time to think too hard about alternatives… the real lock-in is not around the model, but around your environment.” —Eric Paulsen, Coder.

Appealing as all this looks on the surface, will this “promotion” action (and, ah-hem, isn’t that a slightly dirty word to apply to a gleaming frontier model company?) actually push users and developers further down the funnel towards Anthropic?

Eric Paulsen, field CTO for the EMEA region at the secured, self-hosted, agentic AI development environment company Coder, tells The New Stack that he’s seen this kind of pincer movement before.

“These kinds of promotions are a classic top-of-funnel move: get developers and users at large hooked on the workflow before they’ve had time to think too hard about alternatives,” Paulsen says.

And, he thinks, it works.

“Why? Because once you’re deep into a particular provider’s harness and ecosystem, the costs to switch are too high; lifting and shifting your workflow is a project in itself. So the real lock-in is not around the model, but around your environment,” Paulsen says.

Why is AI portability so painful?

Paulsen references the cost to switch, and we know that migrating context between different models is painful, or nigh on impossible. But why is this the case?

This can result from tokenization incompatibility, as each large language model uses its own distinct tokenizer, chunking logic (to break information into smaller parts), and vocabulary size. There’s also the fact that there is no universal AI memory layer; i.e., as DevOps engineer Daniel Ginês has explained, the AI memory portability problem means user context is trapped.

“When you switch, you don’t lose the data – you lose the understanding,” stated Ginês

“Frontier models are today chasing the complexity of tasks in the benchmarking fight to prove their positions. But in reality, within the enterprise, the real complexity is in data, and that is where these models trip badly,” Dheeraj Pandey, DevRev.

The frontier model land grab continues

Part of this promo move by Anthropic may be down to the sheer number of mainstream television news reports now surfacing on how frontier model companies are operating and evangelizing. Perhaps the thinking is… when non-technical laypersons are in the line of fire for this kind of publicity, target the laypersons. That’s how you get lock-in, right?

Founder and CEO of enterprise AI and customer service platform company DevRev, Dheeraj Pandey tells The New Stack that lock-in only works (for Claude Cowork, or indeed Claude itself) if the software tools on offer “become precise, efficient, and safe” for long-term use.

“Frontier models are today chasing the complexity of tasks in the benchmarking fight to prove their positions. But in reality, within the enterprise, the real complexity is in data, and that is where these models trip badly,” says Pandey. “Generating code was a good universal goal for frontier models, but it was also the easiest goal. For co-work to truly work, frontier models need to get a lot more accurate on data retrieval, and GPUs alone can’t get us there.”

Pandey further explains that simple tasks that run on highly complex data continue (in his view) to “yield bad answers”, and he thinks this “over-reliance on MCP is not helping” us as an industry at large. 

Double the usage, double the risk?

In a now-overlaid blog post that heralded the initial launch of Claude Cowork, Anthropic warned of the risks of the tool when fed vague, ambiguous, or contradictory directions and instructions. Accidental file deletion and the potential for prompt injection were cited as possible 

“These risks aren’t new with Cowork,” stated Anthropic at the time. “But it might be the first time you’re using a more advanced tool that moves beyond a simple conversation.”

Six months on, Anthropic hasn’t been overgenerous in reassuring users about such fragilities, but does state that Claude Cowork is “designed with human oversight in mind” and, as it completes tasks, “consequential decisions remain with the user” going forward.

“Anthropic’s approach to agent safety, including how we think about trust, access, and control, is documented in our research,” stated the company.

After the promo party is over 

After July 5, 2026, 5-hour usage limits in Cowork return to their standard levels. There’s no change to user plans or billing.

Claude Cowork is available on all paid plans through the Claude desktop app.