Claude Cowork Shifts to Remote Execution for Fable 5 Agents

Anthropic shifted its Mythos-class models into production availability on July 16, 2026, transitioning Claude Fable 5 from an included subscription benefit to a pure usage-credit pricing model. Detailed in their release on working with Claude Fable 5 in Claude Cowork, the update also expands the agentic workspace beyond local desktop applications into web and mobile beta platforms. For developers and enterprise teams, this marks a structural change in how autonomous tasks are billed and executed.

Fable 5 Specifications and Pricing

Claude Fable 5 is the generally available version of the Mythos 5 architecture. It is optimized for long-horizon tasks, which require autonomous planning, delegation, and self-correction over extended periods.

The temporary free access period that allowed Pro, Max, Team, and Enterprise subscribers to use Fable 5 for up to 50% of their usage has concluded. Fable 5 now operates strictly on usage credits across all Claude surfaces.

Feature	Specification
Intelligence Class	Mythos 5
Context Window	1,000,000 tokens
Output Limit	128,000 tokens
Input Price	$10.00 per 1M tokens
Output Price	$50.00 per 1M tokens

Remote Sessions and Dispatch

Claude Cowork serves as Anthropic’s agentic tool for knowledge workers, designed to navigate file systems, interact with applications like HubSpot or QuickBooks, and perform complex research. Previously restricted to local compute and file access on macOS and Linux, the workspace now uses remote sessions hosted on Anthropic’s cloud infrastructure.

This architectural shift allows tasks to continue running independently even when the user’s local device loses internet connectivity. A new Dispatch feature lets users trigger tasks on their home or office desktops directly from their mobile devices. If you plan to automate desktop workflows, this decoupling of the control interface from the execution environment fundamentally simplifies remote operations. To support the launch, Anthropic has doubled the Cowork session limits for Max plan subscribers through August 5, 2026.

Safety Classifiers and Fallback Mechanics

Fable 5 includes aggressive safety classifiers specifically tuned for cybersecurity and biological workflows. The community response to the model indicates these guardrails are notably more restrictive than previous generations.

When a request flags these safety classifiers, the system automatically triggers a fallback to Claude Opus 4.8 or the user’s preferred secondary model. This ensures the workflow is not interrupted. Users are only billed the lower Opus rates for the fallback response. Anthropic is concurrently running a HackerOne program to refine these classifiers and reduce false positives.

Long-Horizon Execution

The Mythos architecture targets the Software Development Life Cycle and complex compliance audits. In these scenarios, the model operates in an agent harness for days, evaluating its own output and spinning up sub-agents to handle specialized tasks. Understanding how multi-agent systems coordinate is critical when deploying models with this level of autonomy.

The transition to a pure usage-credit model means you must monitor token consumption closely during extended agentic loops. Configure your preferred fallback models in Cowork to ensure aggressive safety filters do not silently stall critical remote sessions while you are away from the interface.