Copilot retires Claude Opus 5, Opus 4.8 and GPT-5.6 Terra on November 16: what changes and how to keep using them
On November 16, 2026, GitHub removes Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra from every Copilot experience, from Chat and agent mode to code completions. GitHub suggests Claude Opus 5.5 for both Opus models and GPT-6.1 Sol or GPT-6 Luna for Terra, and individual subscribers on annual plans keep Opus 4.8. Per token, none of the replacements costs more AI credits than the model it replaces. To keep using Opus 5 or Terra in Copilot, add it with your own key, for example as a Custom Endpoint model in VS Code; Opus 4.8 stays on Anthropic's API.
[
{
"name": "Standard Compute",
"vendor": "customendpoint",
"apiKey": "${input:standardComputeKey}",
"apiType": "chat-completions",
"models": [
{
"id": "anthropic/claude-opus-5",
"name": "Claude Opus 5",
"url": "https://api.stdcmpt.com/v1",
"toolCalling": true,
"vision": false,
"maxInputTokens": 200000,
"maxOutputTokens": 32000
},
{
"id": "openai/gpt-5.6-terra",
"name": "GPT-5.6 Terra",
"url": "https://api.stdcmpt.com/v1",
"toolCalling": true,
"vision": false,
"maxInputTokens": 200000,
"maxOutputTokens": 32000
}
]
}
]
VS Code adds /chat/completions to the URL. For the router, use the id StandardCompute; for the successor, anthropic/claude-opus-5.5.
What GitHub is retiring
| Model | Leaves Copilot | GitHub's suggested alternative | Who keeps it |
|---|---|---|---|
| Claude Opus 4.8 | November 16, 2026 | Claude Opus 5.5 | Individual subscribers on annual plans, "so they retain an Opus offering" |
| Claude Opus 5 | November 16, 2026 | Claude Opus 5.5 | Nobody |
| GPT-5.6 Terra | November 16, 2026 | GPT-6.1 Sol or GPT-6 Luna | Nobody |
Source: GitHub changelog, October 15, 2026, checked October 16. The retirement covers Copilot Chat, inline edits, ask and agent modes and code completions. It is a Copilot change only: Anthropic lists Opus 5 and Opus 4.8 as active on its API, and OpenAI has not deprecated Terra.
Who is affected
- Copilot Pro+, Max, Business and Enterprise users who pick Opus 4.8 or Opus 5. Copilot Pro ($10) doesn't include the Opus models.
- Everyone on a paid plan who picks GPT-5.6 Terra, Pro included. On Pro, GPT-6 Luna is the only one of the two suggested alternatives: GPT-6.1 Sol needs Pro+ or higher.
- Business and Enterprise: under default model enablement the alternatives switch on automatically, unless an admin turned off the global default or disabled the model. Then an admin enables them under the model policies in Copilot settings.
- Annual Pro and Pro+ subscribers still on request-based billing keep Opus 4.8 at its multiplier of 27 premium requests. GitHub says legacy annual plans don't get new models, which is why Opus 5.5 is not offered to them.
- Not affected: Auto model selection, which picks from GPT-5.6 Luna, GPT-5.6 Sol, Claude Sonnet 5 and 5.5, Gemini 3.8 Flash and MAI-Code-1.1-Flash. Nothing needs removing: the retired models disappear from the picker on their own.
What the replacements cost in Copilot
Premium request multipliers no longer apply to monthly plans. Since June 1, 2026, Copilot bills chat, agent mode and CLI use in AI credits by the token (1 credit = $0.01), and completions use no credits on paid plans. Multipliers survive only on legacy annual plans, which is why GitHub publishes no multiplier for Opus 5.5, GPT-6.1 Sol or GPT-6 Luna.
| Model | Input | Cached input | Output | One agent turn |
|---|---|---|---|---|
| Claude Opus 4.8 (retiring) | $5.00 | $0.50 | $25.00 | 11.5 credits |
| Claude Opus 5 (retiring) | $5.00 | $0.50 | $25.00 | 11.5 credits |
| Claude Opus 5.5 | $4.00 | $0.20 | $20.00 | 7.6 credits |
| GPT-5.6 Terra (retiring) | $2.00 | $0.20 | $12.00 | 4.7 credits |
| GPT-6.1 Sol | $2.00 | $0.10 | $10.00 | 3.8 credits |
| GPT-6 Luna | $0.10 | $0.01 | $0.50 | 0.23 credits |
Per 1M tokens, from GitHub's models and pricing page, checked October 16, 2026; prompts over 272K tokens cost more on the GPT models. One agent turn is our average coding-agent turn (90K tokens, 87% cache hits, 700 output tokens) at those rates, cache writes left out. Pro includes 1,500 credits a month and Pro+ 7,000.
How to keep using a retired model
Copilot's bring-your-own-key (BYOK) models don't come from GitHub's model list, so a retirement doesn't remove them. They need a provider that still serves the model.
| Route | How | What to know |
|---|---|---|
| VS Code, Custom Endpoint | Manage Language Models > Add Models > Custom Endpoint, then the model ID (config above) | Chat and agents run on your key. Works without a Copilot plan; on Business and Enterprise an admin must allow BYOK |
| Copilot CLI | COPILOT_PROVIDER_BASE_URL, COPILOT_PROVIDER_API_KEY and COPILOT_MODEL | The model needs tool calling and streaming |
| JetBrains, Xcode, Copilot app | Each has its own BYOK settings | Listed by GitHub; our setup docs cover VS Code and the CLI only |
| Enterprise BYOK | Enterprise owners add an OpenAI-compatible key under AI controls > Copilot > Configure custom models | Public preview, needs a Copilot license. We haven't tested our endpoint there |
| Another agent | Cline, OpenCode, Codex or Claude Code with the same key | No Copilot in the path at all |
Sources: GitHub's bring your own key, Copilot CLI BYOK and enterprise custom models pages, checked October 16, 2026. Completions, Copilot cloud agent and code review stay on GitHub's models whichever route you pick.
Where each model stays available
| Model | Standard Compute | Model maker's API |
|---|---|---|
| Claude Opus 5 | anthropic/claude-opus-5, US pin anthropic/claude-opus-5:us | Active; Anthropic retires it no sooner than July 24, 2027 |
| GPT-5.6 Terra | openai/gpt-5.6-terra, pins :eu and :us | Current; not on OpenAI's deprecation list |
| Claude Opus 4.8 | Not served | Active; no sooner than May 28, 2027. Use an Anthropic key, for example with VS Code's built-in Anthropic provider |
Standard Compute IDs from api.stdcmpt.com/v1/models on October 16, 2026. Anthropic dates from its model deprecations page; OpenAI from its deprecations page.
Should you keep the old model?
Usually not. Opus 5.5 lists at $4 / $20 per million tokens against $5 / $25 for Opus 5 and Opus 4.8, and scores 58 on the Artificial Analysis Intelligence Index against 42 for Opus 4.8. GPT-6.1 Sol scores 52 against Terra's 42 and costs less per turn. Keep the old model where a prompt, eval or agent workflow was tuned on it, and move the rest.
If Terra was your everyday model because it was cheap, GPT-6 Luna is the closer fit on price: a twentieth of Terra's input rate, four points lower on the same index.
Why one API key instead of picking a provider per model
- Cost first: send the model
standardcomputeand the router sends each step to the cheapest model that can do it. File reads, small edits and tool calls go to low-cost models, and Claude Fable 5.1 or GPT-6 Astra take the steps that need them. A pinned ID such asanthropic/claude-opus-5always stays on that model. - For every model we vet providers for price, uptime and speed, with automatic fallback when one fails, so you don't sort providers per request.
- Opt-in Compliance rules restrict a key, team or company to provider endpoints in Europe or the US. Region pins such as
openai/gpt-5.6-terra:eudo the same for one model. - We don't train on your prompts or store them.
- Plans start at $19 a month for a $20 compute budget. At the end of the budget requests stop with HTTP 402 until renewal, with no overage.
The catch
Keeping a retired model costs more, not less. Opus 5 and Opus 4.8 are the most expensive models in this change: at list prices our average agent turn costs about 50% more on Opus 5 than on Opus 5.5, and a pinned Opus 5 uses up a Standard Compute budget about that much faster. A BYOK model also covers chat and agents only: completions, the cloud agent and code review stay on Copilot's own models.
Where Standard Compute fits
Standard Compute serves Claude Opus 5 and GPT-5.6 Terra after November 16, next to Opus 5.5, GPT-6 Luna and the open models, on one OpenAI- and Anthropic-compatible key. Add it to Copilot as a Custom Endpoint model or use it in Cline, OpenCode, Codex or Claude Code. Plans start at $19 a month.
FAQ
When does Copilot remove Claude Opus 5, Opus 4.8 and GPT-5.6 Terra?
On November 16, 2026, in every Copilot experience: Chat, inline edits, ask and agent modes and code completions. GitHub announced it on October 15, 2026.
Does Copilot keep Claude Opus 4.8 for anyone?
Yes. Individual Copilot subscribers on annual plans keep Opus 4.8 so they still have an Opus model, at 27 premium requests per prompt under legacy billing. Everyone else moves to Opus 5.5.
What is the premium request multiplier for Claude Opus 5.5 in Copilot?
There is none on monthly plans. Since June 1, 2026 Copilot bills by the token in AI credits: Opus 5.5 is $4 input, $0.20 cached input and $20 output per million tokens, below the $5 / $0.50 / $25 of Opus 5 and Opus 4.8. Multipliers apply only to legacy annual plans, which don't get new models.
Is Anthropic retiring Claude Opus 5 too?
No. This is a Copilot change. Anthropic lists Claude Opus 5 as active on its API with retirement no sooner than July 24, 2027, and Opus 4.8 no sooner than May 28, 2027.
Can I keep using GPT-5.6 Terra in Copilot after November 16?
Yes, through bring-your-own-key. In VS Code add a Custom Endpoint model with the ID openai/gpt-5.6-terra on a provider that serves it, such as Standard Compute; in Copilot CLI set COPILOT_MODEL to the same ID. Requests then go to that provider and bill there instead of in AI credits.
Do I need to do anything before November 16?
Only if you rely on one of the three models. Copilot removes them from the picker without any action, and Business and Enterprise get the alternatives switched on automatically unless an admin changed the default model policy.