GitHub Copilot retires six models on 2026-10-19: the mapping table, what to change, how to verify
GitHub's changelog of 2026-09-18 deprecates six models across all Copilot experiences - Chat, inline edits, ask and agent modes, code completions - on 2026-10-19, each with an official replacement. This is the second wave: the first (2026-09-01) took out the Claude line. The page keeps to what you need before the date: which name disappears, which one replaces it, which line you edit, and how to prove it worked.
Updated 2026-09-20
Four key points
six models, one date
The official sentence: “We will deprecate the following models across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions) on October 19th, 2026”. That is the same day this page writes as 2026-10-19.
six rows, four replacement targets
Straight from the table: Gemini 3.7 Flash → Gemini 3.8 Flash; GPT-5.5 → GPT-5.6 Sol; GPT-5.4 → GPT-5.6 Sol; GPT-5.4 mini → GPT-5.6 Luna; GPT-5 mini → GPT-5.6 Luna; Grok 4.5 → Grok 4.6. Every row carries the 2026-10-19 date.
it is not only the chat picker
The same sentence names Copilot Chat, inline edits, ask and agent modes, and code completions. Pins sitting in agent-mode settings or completions config are invisible if you only open the chat dropdown.
alternatives for Enterprise/Business
Per the notice, suggested alternatives are automatically enabled for Copilot Enterprise and Business under default model enablement - unless an admin turned off the global default or explicitly disabled the model. Then nothing arrives on its own.
What this is
The second wave of in-product model turnover inside GitHub Copilot. Wave one (2026-09-01) removed six rows: Claude Sonnet 4.5/4.6, Opus 4.5/4.6, plus Gemini 3.1 Pro and Raptor Mini, covered on our own page; wave two removes six non-Claude rows on 2026-10-19. Two confusions are worth killing early. First, Copilot's GPT-5.4 is a product display name, not OpenAI's API id gpt-5.4-cyber. That one leaves the OpenAI API on 2026-10-01 with gpt-5.6-cyber as its migration target, and our catalog does not list gpt-5.4-cyber at all (checked 2026-09-20). Second, GitHub will not rewrite pinned names for you: the notice says to update your workflows and integrations before the date. Removing the retired models afterwards is explicitly a no-action item - No action is required to remove the models after they have been deprecated.
What happened
On 2026-09-18 (pubDate 19:00 UTC) GitHub shipped a changelog entry under the copilot label - upcoming-deprecation-of-selected-github-copilot-models-in-mid-october - carrying the six-row table (Model / Deprecation date / Suggested alternative). This page fetched and verified that entry on 2026-09-20; every quoted sentence above comes from that fetch. If GitHub widens the list before mid-October, re-check before quoting these numbers.
Timeline
2026-09-18: GitHub's changelog announces the 2026-10-19 deprecation of six Copilot models, with the suggested-alternative table and the auto-enable clause.
2026-09-19: this page verifies the notice and our catalog: gpt-5.6-sol and gpt-5.6-luna listed with priced calls (107,852 / 76,349), gemini-3.8-flash listed with 0 priced calls in 30 days.
2026-10-19: all six rows go away across the four surfaces; under default enablement the alternatives are already on, admins who turned the global default off must fix the model policy first, and nothing has to be deleted afterwards.
Confirmed vs caution
Officially confirmed
From the notice: the 2026-10-19 date; the four experience surfaces quoted above; the six old-to-new mappings; automatic enablement of suggested alternatives for Enterprise/Business under default model enablement unless an admin disabled the global default or the model; “update your workflows and integrations” is on your side; no cleanup action after deprecation; Enterprise customers with questions go to their account team.
⚠️ Caution
What this page will not say: that enterprises can request to keep the old models (the notice only offers the account team - no documented exemption); that personal plans are unaffected (not stated); any model list beyond these six rows; that GitHub rewrites hardcoded model ids for you; and anything about Grok 4.6 or the retiring display names being callable on QCode - they are outside the measured table. Copilot-side subscription prices and quotas were not checked and are not inferred. These non-claims were checked against the notice captured 2026-09-20.
Same vs differs
Same
Both waves are GitHub-announced deprecations inside the Copilot product: a hard date, a suggested-replacement table, and neither means the model vendors shut the same ids down on their own APIs the same day. OpenAI's 2026-10-01 removal of gpt-5.4-cyber is its own calendar.
Differs
Wave one (2026-09-01) took six rows: Claude Sonnet 4.5/4.6, Opus 4.5/4.6, Gemini 3.1 Pro, Raptor Mini, documented on our first-wave page; wave two (2026-10-19) takes six non-Claude rows toward Gemini 3.8 Flash, GPT-5.6 Sol, GPT-5.6 Luna and Grok 4.6.
What to do
Two audiences. If you stay inside Copilot: switch to the official alternative before 2026-10-19, then run one real request through each of the four surfaces - chat, inline edits, ask/agent mode, completions - because the pins that break are usually the ones buried in agent settings and completions configs, not the chat dropdown you actually look at. Enterprise and Business admins: confirm the suggested alternatives are enabled; under default model enablement they arrive automatically, but anyone who turned off the global default or explicitly disabled a replacement model has to re-enable it in the model policy first. If you move the call to an API instead: put the exact catalog id in the model field (gpt-5.6-sol, gpt-5.6-luna or gemini-3.8-flash on QCode), send one minimal request, take the 200, then watch the usage dashboard. Verification means calls you observed after the change, not a switch you glanced at.
On QCode
Checked against our own catalog on 2026-09-19: the same-name ids gpt-5.6-sol and gpt-5.6-luna are listed with 107,852 and 76,349 priced calls over the last 30 days; gemini-3.8-flash, the replacement, and gemini-3.7-flash, the model Copilot retires, are both listed with 0 priced calls in the same window. The six retiring names are Copilot product labels, not API catalog entries here; Grok 4.6 is absent from the /models list we captured and shows 0 priced calls in the 30-day ledger. This page promises nothing about availability inside Copilot. Unit prices and rate limits come from /pricing and the live catalog.
FAQ
Which parts of Copilot are affected on 2026-10-19?
The official sentence covers all experiences and names four surfaces: Copilot Chat, inline edits, ask and agent modes, and code completions. Checking only the chat model picker misses agent-mode and completions pins.
Do Enterprise/Business admins need to act?
Normally no: the suggested alternatives are automatically enabled under default model enablement. The exception is an admin who turned off the global default or explicitly disabled a replacement model - then someone must enable it in the model policy before the date.
Is this the same as OpenAI removing gpt-5.4-cyber on 2026-10-01?
No. gpt-5.4-cyber is the OpenAI API calendar: removed 2026-10-01, migrate to gpt-5.6-cyber. Copilot's GPT-5.4 → GPT-5.6 Sol swap is in-product only. Two lists, two targets; do not merge them into one sunset table.
How do I verify the switch worked?
Send one real request per surface after switching and confirm the returned model name is the replacement. API callers: take a 200 on the exact new id and then check the usage dashboard. Observed calls count; eyeballing a dropdown does not.
After the date, must I delete the old models from settings?
No: “No action is required to remove the models after they have been deprecated.” They simply disappear on the date; your work is finishing the switch and the verification before it.
If I call these models via an API instead, which line changes?
The model field in the request body: on QCode it is gpt-5.6-sol or gpt-5.6-luna (both listed, with 107,852 and 76,349 priced calls in the last 30 days) or gemini-3.8-flash (listed, 0 priced calls; all three checked 2026-09-19). Grok 4.6 is outside our measured table, so nothing is promised here. Prices: /pricing.
Sources
GitHub changelog, “Upcoming deprecation of selected GitHub Copilot models in mid-October”, https://github.blog/changelog/2026-09-18-upcoming-deprecation-of-selected-github-copilot-models-in-mid-october/ (RSS pubDate 2026-09-18 19:00:00 +0000; fetched and checked 2026-09-20). The OpenAI API-side removal of gpt-5.4-cyber on 2026-10-01 per developers.openai.com/api/docs/deprecations (checked 2026-09-20). Wave one on our site: /github-copilot-claude-deprecation-sep-2026. QCode catalog listing and 30-day priced usage are our own checks.
Switch before the date, verify, then the deadline is boring
Replacements on QCode: gpt-5.6-sol 107,852 and gpt-5.6-luna 76,349 priced calls in 30 days (checked 2026-09-19); gemini-3.8-flash listed, 0. Prices: /pricing.
Related
GitHub Copilot 2026-09-01 Claude deprecation (wave one)
Which Claude models wave one removed, including the annual-plan terms - on that page.
Gemini 3.8 Flash: pricing and access tiers
The same-name replacement in this wave: its price list and the tiers you can actually get.
GPT-5.6 Sol vs Terra vs Luna
Two of this wave's replacements are Sol and Luna; picking among three tiers is this guide.
Not affiliated with GitHub or Microsoft. All GitHub-side dates, mappings and quotations are transcribed from the 2026-09-18 changelog entry as fetched on 2026-09-20; QCode sells API access, does not provide GitHub Copilot, and cannot speak for Copilot's admin policies.