AI Service Stability Comparison
Compare outage counts, outages per week and median recovery time over the last 90 days for ChatGPT, Claude, Gemini, Grok and other AI services and key infrastructure, from official status page records. Updated every 30 minutes.
AI Models & Services
| Service | Recorded | Incidents (90d) | Per week | Major+ | Days with issues | Median recovery | Recent trend |
|---|---|---|---|---|---|---|---|
 OpenAI (ChatGPT) | 53 days | 54 | 7.1 | 8 | 34 | 1.6h | Usual |
 Claude | 90 days | 46 | 3.6 | 26 | 41 | 62m | Usual |
 Gemini | 52 days | 0 | 0 | 0 | 0 | — | Usual |
 Grok | 52 days | 34 | 4.5 | 2 | 26 | 67m | Usual |
 DeepSeek | 52 days | 12 | 1.6 | 4 | 10 | 14m | Usual |
 Kimi | 88 days | 78 | 6.2 | 78 | 24 | 27m | Usual |
 Upstage | 52 days | 0 | 0 | 0 | 0 | — | Usual |
 Cohere | 53 days | 0 | 0 | 0 | 0 | — | Usual |
 Perplexity | 53 days | 1 | 0.1 | 0 | 1 | — | Usual |
 Poe | 53 days | 0 | 0 | 0 | 0 | — | Usual |
 ElevenLabs | 53 days | 26 | 3.4 | 13 | 11 | 58m | Usual |
 Runway | 90 days | 9 | 0.7 | 1 | 9 | 84m | Usual |
 Microsoft Copilot* | 2 daysShort record | — | — | — | 0 | — | — |
 Midjourney* | 2 daysShort record | — | — | — | 0 | — | — |
 MiniMax | 22 daysShort record | 50 | 15.6 | 50 | 12 | 2m | — |
 Black Forest Labs | 90 days | 6 | 0.5 | 3 | 9 | 4.9h | Usual |
 Recraft | 32 days | 1 | 0.2 | 0 | 1 | — | Usual |
 NVIDIA | 90 days | 1 | 0.1 | 1 | 7 | — | Usual |
 Fish Audio* | 2 daysShort record | — | — | — | 0 | — | — |
 TypeSafe* | 2 daysShort record | — | — | — | 0 | — | — |
 Inception | 33 days | 1 | 0.2 | 1 | 1 | — | Usual |
 Poolside | 33 days | 0 | 0 | 0 | 0 | — | Usual |
* No official incident list, so only the recorded period and days with issues are counted, from the status we check every 30 minutes.
AI Platforms
| Service | Recorded | Incidents (90d) | Per week | Major+ | Days with issues | Median recovery | Recent trend |
|---|---|---|---|---|---|---|---|
 Groq | 53 days | 3 | 0.4 | 1 | 2 | 21m | Quiet |
 Replicate | 73 days | 0 | 0 | 0 | 0 | — | Usual |
 Hugging Face* | 2 daysShort record | — | — | — | 0 | — | — |
 OpenRouter | 61 days | 7 | 0.8 | 0 | 11 | 1.6h | Quiet |
 Inference.net | 33 days | 0 | 0 | 0 | 0 | — | Usual |
* No official incident list, so only the recorded period and days with issues are counted, from the status we check every 30 minutes.
Cloud & Infrastructure
| Service | Recorded | Incidents (90d) | Per week | Major+ | Days with issues | Median recovery | Recent trend |
|---|---|---|---|---|---|---|---|
 Cloudflare | 73 days | 107 | 10.2 | 11 | 73 | 1.7h | Usual |
 Supabase | 52 days | 24 | 3.2 | 4 | 29 | 82m | Usual |
 Vercel | 90 days | 18 | 1.4 | 9 | 15 | 50m | Usual |
 Netlify | 90 days | 16 | 1.2 | 5 | 15 | 52m | Usual |
 Upstash | 90 days | 3 | 0.2 | 1 | 3 | 1m | Usual |
 Pinecone | 90 days | 9 | 0.7 | 5 | 10 | 70m | Usual |
 Google Cloud | 52 days | 2 | 0.3 | 2 | 3 | — | Usual |
 Stripe | 90 days | 2 | 0.2 | 1 | 2 | — | Usual |
 AWS | 52 days | 1 | 0.1 | 0 | 2 | — | Usual |
 Azure* | 2 daysShort record | — | — | — | 0 | — | — |
 Oracle Cloud | 90 days | 0 | 0 | 0 | 0 | — | Usual |
* No official incident list, so only the recorded period and days with issues are counted, from the status we check every 30 minutes.
Developer Tools
| Service | Recorded | Incidents (90d) | Per week | Major+ | Days with issues | Median recovery | Recent trend |
|---|---|---|---|---|---|---|---|
 GitHub | 90 days | 65 | 5.1 | 24 | 44 | 75m | Spiking |
 GitHub Copilot | 90 days | 13 | 1 | 5 | 14 | 64m | Usual |
 Cursor | 75 days | 38 | 3.5 | 5 | 32 | 54m | Usual |
 NPM | 90 days | 10 | 0.8 | 2 | 8 | 1.7h | Usual |
When incidents happen most often
Start times of 329 incidents on AI services over the last 90 days, counted in UTC. The counts are small and can shift by chance, so this is reference only, not a prediction of future incidents. Providers announce incidents differently, so a start time may be when the notice was posted.
Most incidents started during 15:00–18:00, and on Wednesday.
Reading notes
- For services recorded for fewer than 90 days, only incidents in that period are counted. When record periods differ, compare incidents per week rather than total incidents.
- Providers that split incidents into smaller notices show higher counts. A difference in counts can reflect how incidents are announced, not how stable a service is.
- Recovery time runs from the start of an incident until the provider marked it resolved on its official status page. It can differ from how long users were actually affected.
- Infrastructure services such as AWS and Azure span many regions and products, so a single incident often affects only some regions or products.