AI Checker Hub
Archive of significant reliability events and rolling monthly uptime snapshots. Use this page for post-incident review, provider trend comparisons, and planning fallback policy thresholds.
| Date | Provider Scope | Duration | Summary |
|---|---|---|---|
| 2026-02-18 | Multi-provider | 1h 55m | 5xx burst + timeout amplification during high-load period. |
| 2026-01-28 | 2 providers | 48m | Regional rate-limit spikes in EU inference endpoints. |
| 2025-12-11 | Single provider | 2h 12m | Inference queue saturation and elevated latency. |
| 2025-11-03 | Single provider | 37m | Authentication edge instability for API key validation. |
| Provider | Last 30d | Last 90d | Trend |
|---|---|---|---|
| OpenAI | 99.86% | 99.71% | Stable |
| Anthropic | 99.79% | 99.63% | Stable |
| Google AI | 99.41% | 99.18% | Improving |
| Mistral | 99.74% | 99.52% | Stable |
| Cohere | 99.57% | 99.34% | Stable |
| Perplexity | 98.92% | 98.47% | Volatile |
Historical incidents are most useful when they change operational thresholds. If recurring windows show p95 growth before 5xx spikes, move alerting earlier to tail latency and timeout growth rather than waiting for hard failures. This reduces user-visible impact and shortens mitigation time.
Archive data also helps calibrate escalation rules. Short, self-healing bursts should trigger local mitigation, while multi-interval sustained errors should trigger incident mode and fallback routing with explicit traffic caps.
For deeper operational analysis, review the full incident pages: Incident Analysis, OpenAI Status, Anthropic Status, Gemini Status.
Archive review should directly improve your monitoring configuration. If the same event type appears repeatedly, your alert thresholds or escalation logic likely need adjustment. The goal is not to collect history, but to reduce future incident impact and response time.
At least 90 days for routing policy decisions and 12 months for annual reliability planning.
No. Archive trends guide policy; live telemetry drives immediate response.
Weekly for active operations teams and monthly for threshold and runbook updates.
Usually no. Look for repeated patterns and cross-region impact before making major strategy changes.
Use both signals, then prioritize your own production impact and user-facing telemetry.
Quarterly reviews are where archive data creates long-term value. Instead of reading incidents one by one, group them by impact type, affected region, and recovery speed. This reveals whether your platform risk is shifting toward latency, authentication, or hard outages.
Converting answers into explicit roadmap items helps prevent repeated incident patterns and improves operational maturity over time.
Historical status data can guide where engineering time produces the biggest reliability gains. Start by estimating which incident pattern caused the highest user impact and operational cost. Then prioritize one investment at a time: better alerting, safer retries, stronger fallback coverage, or improved auth controls.
This method keeps roadmap choices evidence-based and reduces the chance of chasing low-impact reliability work.
Use the same scoring model each quarter so trend comparisons remain consistent and actionable.