Copilot’s GPT-6 Luna falters again, and GitHub points users to Fusion
GitHub reports degraded availability for GPT-6 Luna in Copilot products and IDEs, blames an upstream provider, and advises switching models or choosing Fusion.

GitHub, owned by Microsoft, is reporting degraded availability for the GPT-6 Luna model in Copilot, across Copilot products and IDE surfaces. The company traces the problem to an upstream model provider and suggests users pick another model or select Fusion.
What GitHub has said so far
The first status entry went up at 15:56 UTC on November 19, saying GitHub was investigating reports of degraded performance for Copilot AI Model Providers. A minute later GitHub confirmed the degradation for GPT-6 Luna and named the cause as an issue with an upstream model provider. It did not name that provider.

While it works with the provider, GitHub recommends choosing another model or selecting ‘Fusion’ to keep using Copilot. The page lists one affected component, Copilot AI Model Providers. At the time of writing it gives no error rates, no estimate for a fix and no count of affected users. The incident is still open.
A pattern that is now hard to ignore
This is not the first time Luna has been caught in a provider problem. On November 13, GitHub reported degraded availability for four models: GPT-6.1 Terra, GPT-5.6 Sol, its faster variant and GPT-6 Luna. According to that incident’s timeline, users saw elevated error rates, and service returned to normal at 21:27 UTC after the upstream provider applied a mitigation. GitHub’s own summary put the start of the degradation at 19:05 UTC, but the first public entry came at 20:13 UTC. We covered that outage at the time.
On November 9 there was another incident, this time for GPT-5.3 Codex and GPT-5.6 models, and on November 18 GitHub listed a separate incident for MAI Code 2 Flash. Four provider-linked notices in eleven days is a short interval. GitHub has not said whether any of them share a cause, and in the November 9 case a root cause analysis was promised but, as far as we could find, has not appeared.
What developers should do in practice
For anyone who has Luna pinned in the model picker, the practical effect is failed or slow requests until the provider recovers. The workaround is simple: change the model. What matters more is what the alternative is.
On November 13 GitHub’s advice was to choose another model or select ‘Auto’. Today it names Fusion instead. Fusion, described in a GitHub changelog entry from November 17, lets Copilot choose and coordinate models for each task, with the user setting an effort level. According to that post, the Max tier uses Claude Sonnet 5.5, High uses GPT-6.1 Terra and Standard uses MAI-Code-2. GPT-6.1 Terra was itself on the November 13 list of degraded models, so Fusion is not immune to provider trouble in principle, and today’s notice does not say which Fusion tiers are unaffected.
GitHub’s changelog says Fusion is available on github.com, in Visual Studio and in the GitHub Mobile beta apps for iOS and Android, with the production mobile apps coming shortly. It does not say which Copilot plans include it. Teams on plans or in editors where Fusion does not appear will have to fall back on picking a model by hand.
The same changelog reports that Fusion Max reached a 71.2% resolution rate on Terminal-Bench 4.1, 5.4 percentage points ahead of Opus 5.5 and Fable 5.2, at lower cost. These are GitHub’s figures and we have not independently checked them. They show why the company is steering users there, but they say nothing about reliability during an outage. Our earlier piece on the Fusion leaderboard result covers the benchmark in more detail.
What the notice leaves open
The page does not say which upstream provider is involved, how many requests are failing or when service will be restored. It also does not say whether Luna usage is billed during the degradation, or whether Fusion routes around Luna automatically. Those are questions teams relying on Copilot for daily work will want answered, and GitHub’s status feed, which offers email, Slack, webhook and RSS updates, is where any resolution will appear first.
Sign Up for the Daily AI Pulse
One email a day. All the stories that matter.
By signing up, you agree to our Terms of Use and acknowledge the data practices in our Privacy Policy. You may unsubscribe at any time.
Alex CarterAI News & Big Tech CorrespondentAlex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.
You may also like

Copilot’s dashboard now counts chat users, and “passive” becomes “chat-first”

GitHub Copilot joins JetBrains Air, with cloud tasks billed to your Copilot plan

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