ChatGPT vs Claude: 44 vs 17 October Incidents, and No Consumer Uptime Promise Found

Published: November 7, 2026 Impact: High. Affects every paying ChatGPT and Claude subscriber who has hit an error screen and wondered what the plan owes them.

OpenAI's status page logged 44 incident entries in October. Anthropic's status page logged 17, on 11 different days. These are counts of status-page entries, never hours down. The question for a paying subscriber is what the written terms give back.

I found no consumer uptime guarantee from OpenAI or Anthropic, and both terms cap what they owe you at the greater of what you paid or $100.

What Happened (This Week)

OpenAI's 44 October entries include 22 whose titles name ChatGPT, Codex, Work or Dots. The rest cover the API, Ads, FedRAMP, Enterprise admin and help-center entries. OpenAI's own labels for the 44: degraded performance 33, partial outage 6, operational-level 5. None was a full outage. Examples: Oct 22, "Elevated ChatGPT conversation and Responses API errors" (about 392 minutes, created to last resolved update, my reading of the feed). Oct 27, "6.1 Sol Degraded Model Performance" (78 minutes).

November so far: four OpenAI entries, including Nov 5–6 "ChatGPT Work and Dots Errors" (about 787 minutes, roughly 13 hours) and Nov 6 "Elevated error rates with some U.S. business Mastercards", a billing entry still being monitored on Nov 7.

Anthropic: 9 of its 17 October entries name Claude models and are major or critical. Oct 23 was the critical one, "Elevated errors for Claude Fable 5.1, Sonnet 5.5, and Haiku 5.5" (34 minutes). Oct 25 hit Sonnet 5.5 and other models for 99 minutes. Sonnet 5.5 was hit twice on Oct 27 (42 and 41 minutes). None so far in November, as of Nov 7 morning.

What This Means

The OpenAI Terms of Use (also the rest-of-world version, effective January 1, 2026) say: "WE DO NOT WARRANT THAT THE SERVICES WILL BE UNINTERRUPTED, ACCURATE OR ERROR FREE". Liability is capped at the greater of 12 months of payments or $100.

Anthropic's Consumer Terms (effective Oct 8, 2025) disclaim warranties of "AVAILABILITY, RELIABILITY". The cap is the greater of six months of payments or $100.

OpenAI's refund article, as of Nov 7, says: "Payments are generally non-refundable." It adds that OpenAI may also approve refunds for "a documented OpenAI service or access failure". That is discretionary and undefined. It does not say who documents the failure or whether a status-page entry counts.

Anthropic's refund article contains none of the words outage, downtime, unavailable, service failure or interruption. That describes the article, not whether any remedy exists.

SLAs exist for business and API buyers only. OpenAI's Scale Tier page says: "Higher reliability: Scale Tier traffic offers a 99.9% uptime SLA and prioritized compute." It lists no credit schedule. Anthropic's service tiers page says: "Priority Tier targets 99.5 percent uptime with prioritized computational resources." It also says: "Priority Tier capacity commitments are no longer available for purchase." and "Service tiers are legacy."

OpenAI publishes uptime on its status page. For August to November 2026, ChatGPT (16 components) shows 99.70% and APIs (14 components) 99.97%. By my arithmetic, 0.30% of about 90 days is roughly 6.5 hours, averaged across the 16 components. It is not hours fully down. I captured no Anthropic uptime figure and print none.

Consumer Protection Q&A

Q: Does my ChatGPT or Claude plan include an uptime guarantee? A: I found no consumer SLA from either vendor. Guarantees appear only on business and API tiers.

Q: Can I get a refund after an outage? A: OpenAI says it may approve one for "a documented OpenAI service or access failure". That is discretionary. Anthropic's refund article does not mention outages.

Q: Is 44 entries worse than 17? A: Not on its own. OpenAI's count includes API, billing, compliance and help-center entries. Neither count measures hours down.

Q: What do the terms cap? A: Each vendor's liability is capped at the greater of what you paid in the prior period (12 months at OpenAI, six at Anthropic) or $100.

What Happens Next

• 30 days: watch whether the Nov 6 billing entry closes and whether Anthropic's page stays clear.
• 90 days: watch whether either vendor defines what counts as a documented failure, or publishes uptime for Claude consumer plans.
• 6-12 months: watch whether any consumer plan, ChatGPT or Claude, gets a written availability commitment.

What to Do Next

• If you hit an outage on a paid plan, screenshot the error and the status-page entry with the time.
• If you ask OpenAI for a refund, cite the article wording about a documented service or access failure.
• If you depend on a tool for paid work, check the business or API tier terms before relying on it.
• If you compare plans, put availability terms next to price on the Compare page.

Bottom Line

What's real: 44 and 17 are status-page entry counts. Both consumer terms disclaim availability and cap liability. SLAs sit on business and API tiers. What to watch: whether OpenAI defines "documented" in its refund wording. What's next: OneHuman keeps a dated record of outage counts and terms. Join OneHuman (Free or Pro).

Sources

• OpenAI, Status history, counts reproduced November 7, 2026.
• Anthropic, Status history, counts reproduced November 7, 2026.
• OpenAI, Terms of Use and rest-of-world Terms of Use, effective January 1, 2026, read November 7, 2026.
• OpenAI Help Center, How do I request a refund for my ChatGPT subscription?, read November 7, 2026.
• Anthropic, Consumer Terms, effective October 8, 2025, read November 7, 2026.
• Claude Help Center, Request a refund for a paid Claude plan, read November 7, 2026.
• OpenAI, Scale Tier, read November 7, 2026.
• Anthropic, Service tiers, read November 7, 2026.

Verified by OneHuman · November 7, 2026

Independent. AI-assisted. Human-verified. No ads. No affiliates. No investors.