AI Service Outages Sep–Oct 2026: OpenAI and Claude Incident Stats
110 incidents from seven official status pages, 2026-09-17 to 10-17, broken down by vendor, severity and duration, with retry, fallback and failover tactics.
AI API speed test · IP purity check · Service status monitoring
Hear your network'sEcho.
One click measures real latency from your network to 15 AI model APIs and 9 public frontend CDNs. Free, no sign-up required.
Speed test results
First connection: includes DNS and TLS handshakeRound trip: median of 3 requests over a reused connection
cdnping.org
AI model APIs
ServicesFirst connectionRound tripStatus
-
cdnpingGlobal (Cloudflare network)——Pending
-
AGI PlanMulti-model aggregation gateway——Pending
-
OpenAI APIUnited States——Pending
-
Claude APIUnited States——Pending
-
Gemini APIGlobal (Google network)——Pending
-
xAI APIUnited States——Pending
-
Mistral APIEurope——Pending
-
Groq APIUnited States——Pending
-
OpenRouterGlobal (Cloudflare network)——Pending
-
DeepSeek APIMainland China——Pending
-
Zhipu BigModelMainland China——Pending
-
Kimi Open PlatformMainland China——Pending
-
Alibaba Cloud Model StudioMainland China——Pending
-
Volcengine ArkMainland China——Pending
-
SiliconFlowMainland China——Pending
Public frontend CDNs View speed rankings →
ServicesFirst connectionRound tripStatus
-
jsDelivrGlobal multi-CDN——Pending
-
jsDelivr (Fastly)Global (Fastly)——Pending
-
unpkgGlobal (Cloudflare)——Pending
-
cdnjsGlobal (Cloudflare)——Pending
-
npmmirrorMainland China——Pending
-
BootCDNMainland China——Pending
-
StaticfileMainland China——Pending
-
ZstaticGlobal——Pending
-
360 frontend static asset librariesMainland China——Pending
Right now
How AI model APIs are doing right now.
Every 5 minutes, we probe each vendor's official API endpoint from Cloudflare edge nodes and sync their official status pages.
Probed from Cloudflare edge DFW · updated 4 min ago (10/17 19:35)
cdnpingNormal
79 msMedian latency100%24h reachable
AGI PlanNormal
759 msMedian latency100%24h reachable
OpenAI APINormal
78 msMedian latency100%24h reachable
Claude APINormal
74 msMedian latency100%24h reachable
Gemini APINormal
65 msMedian latency100%24h reachable
xAI APINormal
44 msMedian latency100%24h reachable
Mistral APINormal
167 msMedian latency100%24h reachable
Groq APINormal
90 msMedian latency100%24h reachable
OpenRouterNormal
64 msMedian latency100%24h reachable
DeepSeek APINormal
360 msMedian latency100%24h reachable
Zhipu BigModelNormal
498 msMedian latency100%24h reachable
Kimi Open PlatformNormal
427 msMedian latency94.44%24h reachable
Alibaba Cloud Model StudioNormal
687 msMedian latency100%24h reachable
Volcengine ArkNormal
689 msMedian latency100%24h reachable
SiliconFlowNormal
752 msMedian latency100%24h reachable
Real-world experience from mainland China
The same API can perform very differently across carriers.
The table below summarizes speed test results from visitors in mainland China over the past 7 days, grouped by carrier. More samples mean more accurate results.
| API | All | China Unicom | Mobile |
|---|---|---|---|
| cdnping10 samples | 1221 ms成功 100% | 1388 ms成功 100% | Insufficient samples |
| AGI Plan9 samples | 1869 ms成功 100% | 1966 ms成功 100% | Insufficient samples |
| OpenAI API11 samples | 300 ms成功 9% | 300 ms成功 11% | Insufficient samples |
| Claude API11 samples | 329 ms成功 100% | 342 ms成功 100% | Insufficient samples |
| Gemini API11 samples | 102 ms成功 9% | 102 ms成功 11% | Insufficient samples |
| xAI API11 samples | 269 ms成功 9% | 269 ms成功 11% | Insufficient samples |
| Mistral API11 samples | —成功 0% | —成功 0% | Insufficient samples |
| Groq API11 samples | 292 ms成功 100% | 301 ms成功 100% | Insufficient samples |
| OpenRouter11 samples | 306 ms成功 100% | 527 ms成功 100% | Insufficient samples |
| DeepSeek API11 samples | 120 ms成功 100% | 138 ms成功 100% | Insufficient samples |
| Zhipu BigModel11 samples | 72 ms成功 100% | 190 ms成功 100% | Insufficient samples |
| Kimi Open Platform11 samples | 111 ms成功 100% | 118 ms成功 100% | Insufficient samples |
| Alibaba Cloud Model Studio11 samples | 44 ms成功 100% | 43 ms成功 100% | Insufficient samples |
| Volcengine Ark11 samples | 68 ms成功 100% | 94 ms成功 100% | Insufficient samples |
| SiliconFlow11 samples | 65 ms成功 100% | 65 ms成功 100% | Insufficient samples |
Values are the median round-trip time over a reused connection. “Success” is the share of requests that received a response within 6 seconds.
New · IP & network environment check
Is your exit IP clean?
Native vs. broadcast, datacenter IP, proxy flags, DNS and WebRTC leaks, split routing rules, browser fingerprint consistency: 12 checks in one go, all run locally.
Blog
Latest articles
AI API Latency Breakdown: Connect Time, Gateway RTT, and TTFT
Why is your AI API slow? Split a call into connect time, gateway round trip, and time to first token (TTFT), with curl commands to measure each segment.
Browser Fingerprint: What Timezone, Language and IP Mismatches Reveal
Sites read your timezone, language and fonts, which often clash with your proxy exit IP. See what each mismatch reveals and how to check it with cdnping /ip.
Claude Code Connection Timeout and Errors: 4-Layer Checklist
Fix Claude Code connection timeouts and errors layer by layer: service status, local network, exit IP and local config. Includes a curl timing command, an error-code table and cdnping tools.
Cloudflare for AI Apps: Caching, Rate Limiting and Origin Protection
What to cache, how to rate limit AI APIs, lock your origin to Cloudflare, and fix SSE buffering and 524 errors. Includes Workers code and a launch checklist.
Codex Usage Limit Resets: Timeline and How to Track the Next
A sourced timeline of the two 2026 Codex usage limit resets, how short-window limits differ from weekly limits, and how to track the next reset with cdnping.
Every Monday morning
One email to catch up on last week's AI APIs and CDNs.
Availability, latency changes, official incidents, and measurements across Chinese carriers, automatically compiled into a weekly report.
FAQ
About cdnping
What does cdnping measure?
Sends requests straight from your browser to 15 AI model APIs and 9 public frontend CDNs, recording first connection time (DNS, TCP, and TLS included) and round trip time over a reused connection. It measures what these services actually feel like from your network, not theoretical numbers between data centers.
What's the difference between "first connection" and "round trip"?
First connection includes DNS resolution and the TLS handshake, showing how long it takes to open a new connection. Round trip is the time for a follow-up request once the connection is established, which is closer to the fixed overhead of each API call. Both need to be low for a good experience.
Why do some APIs show as unreachable?
Some overseas services can't be reached directly from certain networks, or may be blocked by ISPs or firewalls. cdnping records the results as they are, which is valuable information in itself. A failed test doesn't mean the service itself is down; compare it with the edge probe data on the Model API status page to tell.
Does the speed test collect my information?
When "Contribute anonymously to public stats" is checked, we store only the timing and success of each target, plus the country/region and carrier network number (ASN) determined by Cloudflare. We do not store IP addresses or use cookies for tracking. If unchecked, results are shown only in your browser.
How often is the data updated?
Edge probes run every 5 minutes; visitor speed test data is aggregated in real time; official status pages sync every 15 minutes; and the previous week's stability report is generated automatically every Monday.