Anthropic vs OpenAI
Independent checks, recorded incidents and SLA terms for these vendors.
30-day window ending
Independent checks
Probes check endpoint responses. A passing check does not confirm that a customer action worked.
| Last 30 days | Anthropic | OpenAI |
|---|---|---|
| Passing checks | 100.000% 2,880 conclusive checks | 100.000% 2,877 conclusive checks |
| Response time95th percentile | 180 ms | 327 ms |
Response time uses successful checks only: 95% returned within this duration. Different endpoints and regions can produce different results.
Anthropic · check scope
apiRegion not recorded
api.anthropic.com
- First sample in this window
- Latest recorded sample
OpenAI · check scope
apiRegion not recorded
api.openai.com
- First sample in this window
- Latest recorded sample
3 inconclusive checks excluded from the percentage.
View OpenAI recordWhat the incident records say
Recorded starts in the same 30-day window, excluding maintenance. Vendors report incidents differently; fewer posts don’t prove better reliability.
| Last 30 days | Anthropic | OpenAI |
|---|---|---|
| Incidents recorded | 16 | 1 |
| Recorded incident durationMedian | 32m 16 valid durations | Too few dated records 1 valid duration |
Duration uses at least 5 incidents with vendor-supplied starts and recorded closure times. First-observed starts are excluded. A record may close when a post disappears or our tracking period expires; this does not confirm recovery. A missing closure time does not mean an outage is ongoing.
Anthropic
Intermittent error spikes for Claude Mythos 5.1 and Claude Fable 5.1
- Started
- Closed in record
Elevated errors for Claude Mythos 5.1 and Claude Fable 5.1
- Started
- Closed in record
Elevated latency on the Claude API for some users
- Started
- Closed in record
OpenAI
Investigating unexpected usage limit resets
- Started
- Closed in record
Check the terms for your plan
An SLA applies to specific products and plans. Credit eligibility also depends on exclusions, evidence and filing deadlines.