Claude Cowork Expands to Web and Mobile — and Keeps Working When You Close the Laptop
Anthropic’s agentic Claude Cowork — a desktop-only app since January — is now available on web and mobile. The bigger shift is under the hood: sessions now run remotely, so files and progress sync to your Claude account across devices and scheduled tasks keep running even with no device online. Start a task at your desk, check status from your phone, and pick up the finished output later, even after the laptop lid is down.
The rollout is in beta, starting with the Max plan and expanding to more plans over the next several weeks. Anthropic also added Microsoft 365 write tools, so Cowork can draft and send email, manage calendars, and create or update OneDrive and SharePoint files — pushing it further past “coding agent” into general knowledge work.
Fable 5 Is Now Usage-Credit Only at $10/$50 per Million Tokens
Yesterday, July 7, was the last day Claude Fable 5 was included in Pro, Max, Team, and select Enterprise plans at no extra cost. As of today, reaching Anthropic’s most powerful model requires usage credits at the standard API rate of $10 per million input tokens and $50 per million output tokens — roughly double the $5/$25 rate of Claude Opus 4.8.
The practical read: if your team leaned on Fable 5 during the included window, make sure usage credits are enabled or access simply stops. For routine work, Opus 4.8 and Sonnet 5 remain the cheaper default — reserve Fable 5 for the tasks that genuinely need its extra capability, because every call now lands on the bill.
Claude for Government Desktop Enters Public Beta with Claude Code and Cowork
Anthropic launched a public beta of Claude Code and Claude Cowork inside Claude for Government Desktop, built on a FedRAMP High authorized environment. Agency staff can now delegate file-based work — memo drafting, RFP reviews, casework, and decks — to Claude working directly on desktop files, the same cross-device Cowork capability landing for commercial users this week.
The government build layers on the controls public-sector IT expects: stronger admin controls, tamper-evident audit logs, and spend governance for agencies. It’s a clear signal that Anthropic sees regulated and public-sector deployment — not just developer adoption — as a front line for agentic AI.
Claude Code Ships Login-Expiry Warnings and Steadier Background Sessions
The latest Claude Code updates focus on the rough edges of long-running agentic work: login-expiry warnings so a session doesn’t silently die mid-task, clearer agent status and manual-mode badges, and smoother background sessions with macOS, Windows, and daemon fixes. Streaming responsiveness improved and startup memory is trimmed.
None of it is flashy, but it’s the connective tissue that makes background and remote agents usable. As sessions move off the local machine and run for hours, the difference between a helpful agent and an abandoned one is whether it tells you when it needs attention — exactly what these fixes target.
Anthropic’s Own Data: 90% of Cowork Sessions Aren’t Software Development
Alongside the mobile launch, Anthropic published an analysis of 1.2 million anonymized Cowork sessions (sampled May 11–31 across 600,000+ organizations) — and the picture is decisively non-technical. The top category is business process and operations at 33.4% (reports, onboarding checklists, spreadsheet reconciliation), followed by content creation and copywriting at 16.4%. Software development? Just 8.7%.
Anthropic calls these dominant tasks “the work around the work” — the synthesis and coordination that spans nearly every role but rarely appears in a job description. Rounding out the mix: DevOps 7%, research 6.4%, data analysis 5.8%, document processing 4.1%, and sales ops 4%. It reframes Cowork as an office agent that happens to code, not a coding agent that wandered into the office.
Claude Science Connects Natively to NVIDIA BioNeMo
Claude Science, the research workbench Anthropic launched June 30, now connects natively to the NVIDIA BioNeMo Agent Toolkit, exposing high-performance computing resources as callable skills inside the Claude environment. For researchers, that means orchestrating GPU-accelerated genomics, proteomics, and drug-discovery workloads from the same coordinating agent that fronts Claude Science’s 60+ curated scientific skills.
It’s a concrete example of the “skills as compute” pattern: instead of wiring up HPC pipelines by hand, a scientist asks the agent, and the heavy lifting routes to BioNeMo behind the scenes. The AI for Science grant applications tied to the beta remain open through July 15.
Cowork’s Cross-Device Move Is a Bet on “the Work Around the Work”
Put today’s two Cowork stories side by side and the strategy snaps into focus. The usage data says 90% of what people do in Cowork isn’t coding — it’s reports, checklists, decks, and reconciliation, the connective tasks that span every role. The product move answers that data directly: take the agent off the developer’s laptop, put it on a phone and the web, let it run in the background, and hand it Microsoft 365 write access. Anthropic isn’t chasing coders here; it’s chasing the operations manager who’d never open a terminal.
That repositioning has a cost attached, and it’s the day’s quieter story: Fable 5 just moved to usage-credit-only pricing at double Opus’s rate. The pattern for teams is now two-sided — capability is getting more accessible (any device, any role) while the frontier is getting more expensive (metered by the token). The winning play isn’t reaching for the biggest model for everything; it’s letting a cheaper model handle the “work around the work” and saving Fable 5 for the problems that actually earn its price.