GPT-6.1 Terra and three more Copilot models hit errors, blamed on a provider

GitHub Copilot had elevated error rates on several GPT models for more than two hours on November 13, according to the GitHub status page. GitHub blamed an issue with an upstream model provider, and the incident closed at 21:27 UTC.

Four model variants, one upstream cause

GitHub’s final summary says the degradation ran between 19:05 and 21:27 UTC and affected GPT-6.1 Terra, GPT-5.6 Sol and GPT-6 Luna. The 20:47 UTC update also named gpt-5.6-sol-fast and said availability was degraded in Copilot products and IDE surfaces. GitHub did not name the provider.

The status page logged the first public entry, “Investigating”, at 20:13 UTC, and the 20:14 UTC update confirmed degraded availability. That is roughly an hour after the start time in the final summary. The page does not explain the gap. GitHub says its automated monitoring detected the issue and that its automated model-warning system put in-product warnings on the affected models.

The advice to users did not change during the incident. Anyone affected was told to pick another model or choose Auto, which lets Copilot select a model. At 21:17 UTC GitHub said the provider had made mitigations and the models were starting to recover. The incident was resolved ten minutes later. GitHub has not published a root cause beyond “an issue with an upstream model provider”, and it gives no error-rate figures or counts of affected users.

A second GPT incident in the same week

The incident is not the first of its kind this week. On November 9, GitHub reported degraded availability for GPT-5.3 Codex, GPT-5.6 Sol and GPT-5.6 Luna. According to that page, it was marked resolved at 00:05 UTC on November 10, and GitHub said a detailed root cause analysis would be shared when available. The same upstream explanation appears in both incidents, though GitHub does not say whether they are linked.

Both incidents hit the Copilot AI Model Providers component and the GPT Sol and Luna models, which suggests the exposure sits in the dependency rather than in Copilot itself. That is an inference from the two status pages, not something GitHub has stated.

What to watch: Fallbacks and a model lineup that is about to shrink

Copilot’s model picker is a multi-vendor layer, so a failure at one provider shows up as a failure in a model GitHub does not host itself. Auto mode is the practical hedge, since it moves the user off a failing model without a manual change. Teams that pin a specific model in IDE settings or scripts are the ones most exposed to this kind of outage.

The lineup is also changing. In a changelog post the same day, GitHub said it will retire six models across all Copilot experiences on December 14, 2026: Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3. It suggested replacements for five of them and left Kimi K3 as “to be announced”. Claude Haiku 4.5 stays available to individual subscribers on annual plans. Anyone with a hard-coded model name will need to move before that date.

Provider-side trouble has not been limited to GPT. Earlier on November 13, we reported that sign-ins failed on claude.ai and Claude Code, a reminder that coding assistants depend on several vendors at once.

What would settle the open questions: a published root cause for the November 9 incident, any statement on whether the two events share a cause, and whether GitHub explains the gap between the 19:05 UTC start and the 20:13 UTC first entry. Until then, the evidence is limited to GitHub’s own timeline, and no independent measurement of the error rates has been published.