Article

Replace GitHub Copilot models retired on November 16

Copilot users can check replacement models, annual-plan exceptions and saved settings, then compare access and usage costs before switching.

Editorial illustration for Replace GitHub Copilot models retired on November 16: an arrow connects an old service to its replacement. Not documentary evidence.

GitHub retired three Copilot model options across most experiences on November 16, 2026, affecting users and teams whose workflows select them. GitHub recommends Claude Opus 5.5 for Opus 4.8 or Opus 5, and GPT-6.1 Sol or GPT-6 Luna for GPT-5.6 Terra, subject to account access. Individual annual subscribers can still use Opus 4.8.

The change was announced on October 15; the November notice confirms it has taken effect. The task now is to choose an eligible replacement, update saved selections and check its usage costs.

Check your plan before switching

The suggested replacements do not all have the same eligibility. GitHub’s current plan matrix lists:

  • GPT-6 Luna: Pro, Pro+, Max, Business and Enterprise.

  • GPT-6.1 Sol and Claude Opus 5.5: Pro+, Max, Business and Enterprise—not ordinary Pro.

For a usage-based Pro subscriber replacing Terra, that makes Luna the named option available without moving to a higher plan. This is an access distinction, not evidence that Luna and Sol deliver equivalent results. Check availability in your own client; Copilot Free and Student provide automatic model selection only.

The annual-plan exception is important: this retirement does not force individual annual subscribers to replace Opus 4.8. But those on legacy annual request-based billing do not receive new models or features. Their retained Opus 4.8 currently uses a 27× premium-request multiplier, not the token-based credit calculation below. Keeping Opus 4.8 does not imply entitlement to Opus 5.5.

Opus 5 has a different exception. GitHub says it has left the picker but remains available for “runtime refusal fallback” through existing authorized access, with enterprise policies still applying. That limited backend role does not make it a normal selectable replacement or guarantee that an old saved selection will be redirected.

Enable the replacement on workplace accounts

For Business and Enterprise, default enablement normally makes the alternatives available, unless an administrator has disabled the global default or the particular model. A paid seat alone does not override those restrictions.

An organization owner with delegated control can open organization Settings → Copilot → Models and enable the intended replacement. An enterprise-enforced setting must be handled by the enterprise owner. If the enterprise uses the enterprise-teams model-access preview, organization model settings are unavailable; use the enterprise controls instead.

Enterprise owners manage these choices under AI controls → Copilot → Configure models. Check the specific replacement rather than changing the organization’s broader defaults just to make one migration work.

Update Chat and saved model settings

Update saved model choices as well as the Chat picker. In VS Code, open Copilot Chat and select the replacement from the current-model dropdown at the bottom of the chat view. That selection does not change the model used for inline suggestions.

For command-line workflows, run /model inside an interactive Copilot CLI session to see the available model identifiers and save a default. Use those identifiers rather than guessing one from a display name. Then inspect scripts and custom-agent profiles for their own model selections.

A new default can lose to a higher-priority setting. GitHub’s CLI reference gives this order, highest first:

  1. A custom agent’s model when model-policy: required is set.

  2. An explicit --model argument.

  3. An ordinary custom-agent model selection.

  4. The COPILOT_MODEL environment variable.

  5. The model setting in ~/.copilot/settings.json or the settings file under COPILOT_HOME.

  6. The CLI’s built-in default.

Change the setting that actually controls the workflow, preserving unrelated configuration and tool permissions. GitHub does not specify a universal outcome for every old model identifier, so do not assume saved profiles will be rewritten automatically.

Compare usage rates before relying on the replacement

On usage-based plans, tokens—the units of text sent to or generated by a model—determine credit consumption. Cached input is reused context; cache writes store context for reuse. GitHub prices these separately and converts usage at one AI credit per US$0.01.

The following Copilot rates, checked on November 16, are in US dollars per million tokens. Sol and Luna use these default rates for requests with no more than 272,000 input tokens.

Model

Input

Cached input

Cache write

Output

Claude Opus 5.5

$4.00

$0.20

$5.00

$20.00

GPT-6.1 Sol

$2.00

$0.10

$2.50

$10.00

GPT-6 Luna

$0.10

$0.01

$0.125

$0.50

Above 272,000 input tokens, Sol’s rates become $4 input, $0.20 cached input, $5 cache write and $15 output per million. Luna’s become $0.20, $0.02, $0.25 and $0.75 respectively. The rate card lists no separate long-context tier for Opus 5.5.

Illustrative calculation: assume one call uses 100,000 ordinary input tokens and 10,000 output tokens, with no cache reads or writes, discounts or additional calls. Multiplying the input rate by 0.1 and the output rate by 0.01 gives:

  • Opus 5.5: $0.40 + $0.20 = $0.60, or 60 credits.

  • Sol: $0.20 + $0.10 = $0.30, or 30 credits.

  • Luna: $0.01 + $0.005 = $0.015, or 1.5 credits.

These are equal-token arithmetic examples, not measured tasks or quality comparisons. An agent task can make several model calls, and different output lengths or cache usage change the total. Usage drawn from an included allowance is not necessarily an extra bill. A lower rate alone cannot establish the cheapest model for a correctly completed task.

For Business and Enterprise, credits come from a shared pool and additional paid usage is enabled by default. To prevent overages, review the AI credits paid usage policy. Do not confuse that with a $0 user-level budget: the latter blocks credit-consuming features immediately, including access to the included pool.

If you are comparing options beyond GitHub’s named replacements, see Gemini 4 Hydrogen access and billing in Copilot. Hydrogen is not one of the suggested alternatives in this November notice.

Confirm the change and check the next deadline

Before relying on the updated workflow, try a small, familiar task and review the response or code changes and recorded usage. Confirm the intended model in the picker or CLI response output—noninteractive runs show it unless silent output is enabled. This checks your selection and task suitability; it does not establish general equivalence with the retired model.

Finally, check whether you also use any of the six different models covered by GitHub’s December 14 retirement notice. RohitAI’s guide to Copilot’s December model retirements covers that separate deadline. It does not extend the November 16 retirement date.