Detailed Analysis
A Claude Max subscriber using the iOS application has reported a persistent failure state in a feature called Dispatch, in which a server-side thread entered an unrecoverable error loop tied to extended 1M-context usage. The error message presented to the user — "API Error: Usage credits required for 1M context" — indicates the thread was originally initiated under a session configured for maximum context capacity, and that session state has become locked in a condition the client application cannot resolve. The user exhausted all standard self-service remediation steps, including app reinstallation, logout/login cycling, and toggling the Dispatch feature off and on, none of which produced a resolution. Notably, the user confirmed that web access via Safari and Claude Code on Windows continued functioning normally, isolating the failure specifically to the Dispatch thread's server-side state.
The core technical issue the user identifies is architectural: Dispatch, as described, operates via a single persistent server-side thread rather than maintaining any recoverable local client-side state. This design means that when the thread enters a broken state, conventional client-side interventions — reinstalling the app, clearing local data — have no effect, because the state that needs resetting does not reside on the device. The user also notes there is no in-app UI mechanism to delete, reset, or reinitialize a Dispatch thread, leaving no self-service path to resolution and necessitating manual intervention from Anthropic's backend infrastructure team.
The post references a corresponding GitHub issue filed against the anthropics/claude-code repository, suggesting this is a known and documented failure mode rather than an isolated incident. The existence of a tracked issue implies the engineering team is aware of the bug class, but the lack of a self-service resolution path — or an in-app thread management interface — represents a gap in the product's operational resilience. For a paid Max subscriber, the inability to reset a stuck thread without direct support escalation introduces meaningful friction and service degradation.
This report reflects a broader challenge in AI application design as features grow in architectural complexity. Persistent server-side session threads, extended context windows, and tiered credit-based model access introduce multiple interdependent failure surfaces that are difficult to expose cleanly to end users or resolve without backend access. The case illustrates a tension between stateful, high-capability features — such as long-context Dispatch threads — and the user expectation that basic session management, including the ability to start fresh, should always be available as a self-service action. As Anthropic continues expanding Claude's feature surface across iOS, web, and developer tooling, ensuring coherent state management and accessible recovery paths will be increasingly important to user trust and operational quality.
Read original article →