Anthropic reports Claude network failures with elevated errors

• Claude had two incidents in 24 hours: ClaudeDevs' incident note attributed elevated errors and reduced availability to multiple separate network failures that cut serving capacity while traffic was rerouted.
• Capacity was stable but not fully recovered when Anthropic posted its update, according to ClaudeDevs' follow-up.
• Claude Code was visibly affected: theo's Claude Code post showed a major-outage tooltip lasting 25 minutes, while dejavucoder's status screenshot showed claude.ai in major outage during the incident.
• The error shape matters for API users: Anthropic's docs define 529 as an overloaded_error, and theo's post captured the developer-facing impact as Claude Code being mostly down.
• The community response quickly shifted to quota accounting, with vikhyatk asking for rate-limit resets and kimmonismus arguing Anthropic still resists that kind of community remediation.

The Claude Status page now shows all systems operational, but its incident log records a July 29 all-models outage followed by a July 30 many-models error event. Anthropic's API error docs define 529 as temporary API overload caused by high traffic across users, distinct from a user-specific 429 rate limit. BleepingComputer reported users hitting “API Error: 529 Overloaded,” and CNET saw Downdetector reports fall from more than 4,000 to about 150 as recovery progressed.

ClaudeDevs attributed the 24-hour availability hit to “multiple separate network failures” that reduced Anthropic's capacity to serve Claude while some requests failed during traffic rerouting.

The follow-up was careful: capacity restoration had progressed and the service was stable, but Anthropic said it was “not fully in the clear yet.”

The official status log separates the disruption into two incidents, not one long outage.

• Jul 29, elevated errors across all models: investigating at 19:49 UTC, identified at 20:33 UTC, recovery across most models at 21:38 UTC, monitoring at 22:20 UTC, resolved at 22:36 UTC on Claude Status.
• Jul 30, elevated errors across many models: investigating at 05:57 UTC, Opus 5 still unrecovered at 06:26 UTC, Sonnet 5 and Fable 5 errors moving above baseline through 08:06 UTC, “errors across all models” again at 08:33 UTC, baseline at 09:58 UTC, resolved at 10:48 UTC on Claude Status.
• Jul 30, degraded Opus 4.8 performance: investigating at 13:43 UTC, resolved at 14:24 UTC on Claude Status.

That timeline fits ClaudeDevs' wording: separate failures, repeated capacity loss, then rerouting and recovery.

During the outage, BleepingComputer's report said Claude and API-backed tools returned “API Error: 529 Overloaded.”

Anthropic's API error documentation defines 529 as overloaded_error, meaning “the API is temporarily overloaded,” and says 529s can occur when high traffic spans all users. The same doc distinguishes that from 429 rate_limit_error, which means an account has hit a limit.

The docs also say official SDKs retry transient failures with exponential backoff twice by default, honoring retry-after when present.

Claude Code was in the affected surface area, not just claude.ai. The status page screenshot in dejavucoder's post marked claude.ai as a major outage, and theo's Claude Code post showed Claude Code in major outage for 25 minutes.

WesRoth's read captured why the reaction was louder than a normal chatbot wobble: Claude is now part of developer infrastructure for coding, agents, and automation, so a short disruption can pause multiple workflows at once.

Developer chatter moved from “is it down?” to “will usage be credited back?” within minutes. theo's Claude Code post said the outage was inconvenient but forgivable “if we get a reset.”

kimmonismus made the complaint sharper, saying Anthropic “inevitably refuses this kind of community work” around rate resets.

The public Claude Status entries for the July 29 and July 30 incidents describe investigation, recovery, monitoring, and resolution, but do not mention quota or rate-limit resets.

The status page's 90-day view puts the “three nines” joke in context. At read time, Claude Status listed all systems operational with these 90-day uptime figures:

• claude.ai: 99.33%
• Claude Console: 99.79%
• Claude API: 99.40%
• Claude Code: 99.33%
• Claude Cowork: 99.46%
• Claude for Government: 99.99%