ChatGPT, Claude, Grok, and Gemini Hit by Rare Overlapping Outages

Introduction

A cluster of problems affected leading consumer AI services on Thursday morning. Claude, ChatGPT and Codex, Grok, and—according to user reports—Gemini all experienced issues within a relatively short window. Individual model outages are routine, but the overlap among four major services is unusual enough to raise questions about the resilience of the current AI ecosystem.

Key developments

• Anthropic reported elevated request errors for Claude Mythos 5.1, Fable 5.1, and Opus 5 at 9:23 a.m. Eastern. It said the cause had been identified roughly 15 minutes later and marked the incident resolved at 12:16 p.m. A separate, brief increase in errors affected Claude Sonnet 5 shortly after noon.
• OpenAI said at 10:43 a.m. that elevated errors across ChatGPT and Codex were causing degraded performance. After mitigation was deployed, the incident was marked resolved at 12:55 p.m.
• Grok displayed a user-facing message saying the service was experiencing problems. DownDetector reports rose from fewer than 10 shortly before 9 a.m. to 1,365 around 9:45, then declined.
• Google did not publicly acknowledge a Gemini incident. However, DownDetector reports increased from about 23 to 412, while StatusGator classified a Gemini API disruption between 10:45 and 11:15 a.m. as a “likely outage.”

Why the overlap matters

Nothing in the available reports demonstrates that the four incidents shared a root cause. AWS, Microsoft Azure, and Cloudflare did not report major problems at the time, although all three saw some increase in user reports. It would therefore be premature to blame a particular cloud platform, network provider, or common software dependency.

The more durable lesson concerns concentration. Individuals and businesses increasingly use a small number of hosted models for drafting, coding, customer support, and automation. When several providers become unreliable at nearly the same time, switching services may not be a practical fallback. Product teams may need status transparency, provider diversity, caching, graceful degradation, and human procedures for critical workflows—not simply a higher uptime target for one API.

The reported 90-day figures were 99.4 percent availability for Claude, 99.63 percent for ChatGPT, and 100 percent for ChatGPT Codex. Those numbers suggest generally strong service performance, but they do not measure the operational impact of a concentrated outage. For AI builders, resilience testing should include cross-model failover and isolation, not just the uptime of an individual endpoint.

Source: Ars Technica AI