Tom's Guide Claude Hit Its Third Error Spike in 12 Days — a Reliability Pattern Developers Can't Ignore
Anthropic's status page shows elevated Claude errors from 00:50 to 02:10 UTC on September 22 across Mythos 5.1, Fable 5.1, and Opus 5 — the third such incident since September 11. Downdetector logged 1,600+ user reports.
Claude stumbled again overnight. Anthropic’s status page records elevated errors from 00:50 to 02:10 UTC on September 22 — roughly 80 minutes — affecting Mythos 5.1, Fable 5.1, Opus 5, and the prior-generation Fable 5 and Mythos 5. The incident hit claude.ai, the API, Claude Code, and Cowork, and is now marked resolved across all systems. Downdetector logged more than 1,600 user reports during the window.
Three spikes in twelve days
What makes this one notable is the pattern. The same status page shows an earlier elevated-error incident on September 11 (13:57–14:20 UTC) and intermittent errors on September 15 affecting Mythos and Fable 5.1. Three reliability events in twelve days, each touching the current flagship models, is no longer a blip — it is a cadence. Anthropic has not published root causes for any of the three on its public status page.
The September 22 window is also the longest of the three, and the only one to span all three current models plus Cowork, Anthropic’s newer agentic workspace product. For anyone running Claude Code sessions or API-backed agents overnight in US time zones, the timing — late evening Pacific — meant the failure landed squarely in prime working hours for Asia-Pacific developers and in the middle of long-running agent sessions elsewhere.
The developer takeaway
Claude Code’s rise has made Anthropic’s uptime a load-bearing assumption in a lot of development workflows. An 80-minute error window breaks in-flight agent runs, burns through retry budgets, and — for API users with failover logic — triggers the kind of cascading cost that only shows up on the invoice. Three incidents in twelve days is the point at which “add a second provider fallback” stops being paranoia and becomes standard architecture.
Worth watching: whether Anthropic addresses the clustering publicly. The company’s safety and alignment communications are famously thorough; its operational communications about these spikes have so far been limited to the status-page entries themselves. Developers building businesses on the API deserve more than a green “resolved” badge — they need to know whether capacity, a bad deploy, or something structural is behind the pattern.
In the meantime, the engineering response is straightforward and worth implementing this week: route through a model-agnostic layer with automatic failover. Whether that means OpenRouter, a thin proxy with retry-then-fallback logic, or simply keeping API keys for two providers warm, the goal is the same — an 80-minute Claude outage should degrade your product to a backup model, not take it offline. Three spikes in twelve days is the market telling you that single-provider architectures are now a known risk, priced accordingly.