Is GitHub Copilot Down?

THREE FRONTIER MODELS LANDED IN COPILOT IN FOUR DAYS -- AND THE ENABLEMENT POSTURE SPLITS EXACTLY ALONG DATA-RETENTION LINES (2026-09-01 to 2026-09-04, vendor-primary): GitHub added **Claude Fable 5.1 (2026-09-01)**, **Gemini 3.8 Flash (2026-09-03)** and **GPT-6 Astra (2026-09-04)**, each within a day or two of its vendor launch. **THE IMPORTANT DIFFERENCE IS NOT THE MODELS, IT IS WHICH ONES TURN THEMSELVES ON.** For Astra and Gemini 3.8 Flash, GitHub's wording is identical and automatic: '**Under default model enablement, new models are enabled automatically unless an administrator has turned off the global default or explicitly disables this model.**' For Claude Fable 5.1, it is the opposite: '**Copilot Enterprise and Copilot Business plan administrators must enable the Claude Fable 5.1 policy in Copilot settings. The policy is off by default.**' **The stated reason is data retention, not capability:** '**Unlike other Claude models in GitHub Copilot, Claude Fable 5.1 requires data retention by default to operate Anthropic's safety classifiers.**' Zero data retention is available only 'for certain eligible enterprise customers'. **This confirms the dividing line this page identified on 2026-08-26: default-on enablement in enterprise Copilot is gated by licensing and data-retention posture, not by how good the model is.** Anthropic's newest and strongest model is the one that has to be switched on by hand, precisely because its safeguards require retaining data. **AVAILABILITY AND TIERS:** Astra and Fable 5.1 are **Pro+, Max, Business and Enterprise** (note: **not** Pro); Gemini 3.8 Flash reaches **Pro, Pro+, Max, Business and Enterprise**. All three appear across VS Code, Visual Studio, Copilot CLI, the coding/cloud agent, the Copilot app, JetBrains, Xcode and Eclipse; Astra and Fable 5.1 also on github.com and GitHub Mobile. Rollout is gradual in every case. **BILLING:** Astra and Fable 5.1 bill at **provider list pricing** under usage-based billing; Gemini 3.8 Flash bills at **introductory provider pricing through 2026-12-31** -- so the Copilot cost of that model doubles on 2027-01-01 along with Google's rate card. **GitHub's own testing note on Astra is about process rather than output: it 'plans and validates as it goes, batches diagnosis with verification, and independently confirms its results before declaring a task done.'**

A FOUR-MODEL DEPRECATION SLATE DATED 2026-10-02 -- INCLUDING CLAUDE OPUS 4.7 AND TWO GEMINI FLASH VERSIONS (2026-09-03, vendor-primary): GitHub will remove the following '**across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions) on October 2nd, 2026**': **Gemini 3.5 Flash** and **Gemini 3.6 Flash** (both -> Gemini 3.8 Flash), **Kimi K2.7 Code** (-> Kimi K3), and **Claude Opus 4.7** (-> Claude Opus 5). **READ THE GEMINI PAIR TOGETHER WITH THE 3.8 LAUNCH: Google shipped 3.8 Flash on 9/02 and GitHub scheduled the retirement of two older Flash versions the next day.** The Flash line is now turning over faster inside Copilot than most teams re-pin their model choices -- three Flash releases in six weeks upstream, two Flash retirements downstream. **THE ADMIN TRAP GITHUB FLAGS EXPLICITLY:** the suggested alternatives are not automatically usable. '**Copilot Enterprise and Copilot Business administrators may need to enable access to the alternative models through their model policies in Copilot settings.**' So an org that is not paying attention can lose Opus 4.7 on October 2 and find Opus 5 not enabled behind it. Verify the policy is on before the date, not after. **No action is required to remove the models after deprecation, and GitHub directs Enterprise customers with concerns to their account team.**

COPILOT BUSINESS AND ENTERPRISE SIGNUPS REOPEN -- WITH UPFRONT PER-SEAT CHARGING THAT HITS EXISTING CUSTOMERS ON 2026-10-01 (2026-09-03, vendor-primary; the dated flip this page has tracked since 8/28, and it landed two days late): GitHub is '**gradually reopening sign-ups for Copilot Business and Copilot Enterprise customers paying by credit card or PayPal over the next couple of weeks**', alongside '**strengthening account vetting and updating billing experiences**' for those customers. **THE BILLING CHANGE IS THE PART THAT COSTS MONEY AND IT IS NOT LIMITED TO NEW CUSTOMERS.** Two changes apply to card/PayPal Business and Enterprise accounts: '**All new Copilot Business or Copilot Enterprise seat assignments will require payment for each seat before users gain Copilot access**', and '**At the start of the next billing cycle, all Copilot Business and Copilot Enterprise seats assigned will incur an upfront charge. This includes existing customers starting October 1, 2026.**' **So an established org that has been billed in arrears switches to paying for every assigned seat upfront on October 1** -- a working-capital change, not a price change. Included usage 'may be prorated across the month to align with the seat cost proration.' **WHAT IS EXPLICITLY NOT CHANGING:** plan pricing, seat prorations, and the ability to buy usage beyond included allowances. **One trap worth knowing: 'Fully canceling Copilot and returning later may trigger updated account vetting and billing behaviors'** -- so churning out and back in is not a way to keep the old billing treatment. Invoiced enterprise customers are outside the scope of this notice.

ENTERPRISE-MANAGED SETTINGS CAN NOW PIN ANY MODEL AS THE DEFAULT, INCLUDING PER-TEAM (2026-09-02, vendor-primary): GitHub made model defaults in enterprise-managed settings generally available -- '**You can now set your preferred GitHub Copilot model as the default for new conversations through enterprise-managed settings**', and '**customize the default by enterprise team, assigning a different default model based on team membership**' by marking the model key overridable and updating `team-mappings.json`. Users not covered by a team mapping inherit the enterprise default. **WHY THIS MATTERS THIS WEEK SPECIFICALLY:** it lands in the same window as three model additions and a four-model deprecation slate, and it is the control that makes those manageable -- an org can pin Gemini 3.8 Flash for a cost-sensitive team and Fable 5.1 or Astra for a team doing long-horizon work, rather than accepting one global default. **It also partially offsets the default-enablement asymmetry above:** Fable 5.1 will not enable itself, but once an admin turns the policy on, it can be made a team's default rather than something each developer selects. GA on **Copilot Business and Copilot Enterprise**, in the GitHub Copilot app, Copilot CLI and Visual Studio Code.

COPILOT CODE REVIEW CAN NOW APPROVE PULL REQUESTS, AND CONTENT EXCLUSIONS REACHED GA IN THE APP AND CLI (2026-09-01 / 2026-09-02, vendor-primary): two smaller changes that matter for governance. **Copilot code review can now approve pull requests (2026-09-01)** -- following the 2026-08-27 removal of the 300-file / 20,000-LOC cap, the reviewer has gone from commenting to being able to register an approval. **Treat this as a branch-protection question, not a feature:** if your repository counts Copilot's approval toward a required-reviewers rule, an automated reviewer can now satisfy a human gate. Check the setting before assuming it cannot. **Content exclusions are generally available in the Copilot app and Copilot CLI (2026-09-02)**, closing a real gap -- exclusions that already applied in the IDE now also apply on those two surfaces, so files you had excluded from Copilot in VS Code are no longer reachable through the CLI. **The standing double-billing caveat on code review still applies.**

THE COPILOT CODE REVIEW SIZE CAP IS GONE, AND IT NOW REVIEWS BOT-AUTHORED PRs (2026-08-27, vendor changelog): three changes shipped together, and the first is the one that actually changes what the product can do. **(1) THE 300-FILE / 20,000-LINE LIMIT HAS BEEN REMOVED.** Copilot code review previously refused pull requests above **300 files or 20,000 lines of code**; that ceiling is lifted, so large refactors, dependency sweeps and generated-code PRs -- exactly the reviews a human is least willing to do carefully -- are now in scope for the first time. **(2) BOT-AUTHORED PULL REQUESTS ARE NOW REVIEWED**, including PRs opened by **Copilot cloud agent**, when reviews are requested automatically. GitHub is explicit that the cloud agent's PRs now get '**a full agentic review**' rather than the previously limited experience. **This closes a real loop: Copilot's own agent writes the PR, and Copilot reviews it.** Whether you regard that as useful automation or as marking your own homework is a judgement call the changelog does not make for you, and there is no second-model review option. Billing detail worth catching: organizations with the '**Allow members without a Copilot license to use Copilot code review**' policy enabled can **bill that usage directly to the organization**. **(3) RESOLUTION REASONS:** when you resolve a Copilot comment you can now pick **Addressed**, **Won't fix**, or **Incorrect** from a dropdown beside the resolve button -- the first structured signal GitHub collects on whether its review comments were actually right, and the 'Incorrect' option is the one to use if you want that feedback to mean anything. **THE COST WARNING ON THIS PAGE APPLIES DOUBLE NOW:** since 2026-06-01 Copilot code review bills **twice** -- the premium request plus GitHub Actions runtime -- so removing the size cap removes the thing that was implicitly capping your spend on runaway PRs. **A 5,000-file migration PR is now reviewable, and now billable.** The changelog does not state which plans or tiers these three changes apply to.