Verdict, October 20, 2026. OP for most people on usage-based billing. The replacements are already generally available, and the published list price is the same or lower. Mid for Copilot Business and Enterprise teams whose admin turned off default model enablement, or explicitly disabled the replacement, because it stays out of the picker until that policy is on. Mid again for any script, prompt file, or custom agent that still names the old model. OP or Mid has not run these models in Copilot.
GitHub's October 15, 2026 changelog sets the date at November 16, 2026. Anyone who pinned Claude Opus 4.8, Claude Opus 5, or GPT-5.6 Terra has to move. The post says the retirement covers all GitHub Copilot experiences, and it names Copilot Chat, inline edits, ask and agent modes, and code completions. It does not name Copilot CLI or code review. Those two are unstated here. Source: changelog (opens in a new tab).
The switch
| Retiring model | Retirement date | GitHub's suggested replacement | Default input / output now | Replacement input / output |
|---|---|---|---|---|
| Claude Opus 4.8 | 2026-11-16 | Claude Opus 5.5 | $5.00 / $25.00 | $4.00 / $20.00 |
| Claude Opus 5 | 2026-11-16 | Claude Opus 5.5 | $5.00 / $25.00 | $4.00 / $20.00 |
| GPT-5.6 Terra | 2026-11-16 | GPT-6.1 Sol or GPT-6 Luna | $2.00 / $12.00 | Sol $2.00 / $10.00. Luna $0.10 / $0.50 |
USD per 1 million tokens, default tier, checked October 20, 2026. Terra's $2.00 / $12.00 card applies up to 272K input tokens. The Opus rows have one rate. Source: models and pricing (opens in a new tab).
The footnote is only on Claude Opus 4.8: it stays for individual Copilot subscribers on annual plans so they retain an Opus offering, and that deprecation does not apply to them. Opus 5 and GPT-5.6 Terra have no such line. Claude Opus 5.5, GPT-6.1 Sol, and GPT-6 Luna are already GA. The November 16 rows are not in the retirement history table yet. Source: supported models (opens in a new tab).
The list price moves down, or stays flat
PRICE Both retiring Opus models are $5.00 input, $0.50 cached input, $6.25 cache write, and $25.00 output. Opus 5.5 is $4.00 / $0.20 / $5.00 / $20.00. That is 20% less on input, output, and cache write. Cached input is $0.20 against $0.50, which is 60% less, and less than half. The page labels all three Powerful.
PRICE GPT-5.6 Terra is Versatile: $2.00 / $0.20 / $2.50 / $12.00 up to 272K input tokens, then $4.00 / $0.40 / $5.00 / $18.00. GPT-6.1 Sol is Powerful: $2.00 / $0.10 / $2.50 / $10.00 up to 272K, then $4.00 / $0.20 / $5.00 / $15.00. Input matches on both cards. Cache write matches. Cached input is half. Output is $10 instead of $12 on the short card, and $15 instead of $18 on the long card, one sixth less either way.
PRICE GPT-6 Luna is Lightweight: $0.10 / $0.01 / $0.125 / $0.50 up to 272K, then $0.20 / $0.02 / $0.25 / $0.75. Against Terra, input, cached input, and cache write are one twentieth on both cards. Output is one twenty-fourth ($0.50 against $12.00, and $0.75 against $18.00).
GitHub names Sol or Luna and does not assign either to a task. Our reading, not a quality test: GPT-6.1 Sol when Terra was doing heavier coding work, GPT-6 Luna for token-billed volume. A finished-task cost needs token counts this page does not give. See the October 18 price-per-task piece. Rates above: models and pricing (opens in a new tab).
PRICE Completions and next edit suggestions are not billed in AI credits and stay unlimited on paid plans. Overage elsewhere uses these rates, at 1 GitHub AI Credit = $0.01 USD. Terra still leaves completions on November 16, but that switch does not change an AI credit bill. Code review picks a model automatically and does not disclose it, so there is no review pin to retarget.
Annual plans keep Opus 4.8
NERF GitHub moved to usage-based billing on June 1, 2026. Multipliers apply only to Copilot Pro and Copilot Pro+ subscribers on an existing annual plan who stayed on the legacy premium request model. Users on legacy annual plans will not receive access to new models and features. Opus 4.8's multiplier on that page is 27 premium requests, and multipliers can change. Source: annual plan multipliers (opens in a new tab).
GitHub's stated reason for keeping Opus 4.8 is the changelog: so those annual subscribers retain an Opus offering. We read the "no new models" note as why that offering is still 4.8, and why Opus 5.5 may not appear. The multipliers page does not say that. The changelog does not extend the exception to Opus 5 or GPT-5.6 Terra, and it does not limit it to Pro and Pro+. Check the picker.
Admins, and why not to hard-pin a name
NERF Under default model enablement, the alternatives turn on for Copilot Enterprise and Copilot Business unless an admin turned off the global default or explicitly disabled the model. If that default is off, enable them in their model policies. They then show in the Copilot Chat model selector. Nobody has to remove the old models afterward. Source: changelog (opens in a new tab). Unconfigured GA models follow that policy until someone sets them. Source: supported models (opens in a new tab).
NERF Retirement history, read October 20, lists Claude Opus 4.5 and 4.6 retired on 2026-09-01 and Claude Opus 4.7 on 2026-10-02. The suggested alternative on each row is now "Latest Claude Opus model". On August 31 the changelog still named Opus 4.7, 4.8, or 5. Sources: supported models (opens in a new tab), August 31 changelog (opens in a new tab). Opus 5 has been in Copilot since July 24, 2026, 115 days before November 16. The history table does not support an 11-week lifespan. Source: July 24 changelog (opens in a new tab). A separate pricing row, Claude Opus 4.8 (fast mode) (preview), is $10.00 / $50.00. The October 15 post does not name it, so this article does not count fast mode as retired.
NEW Gemini 4 Argon is another picker option, not a suggested replacement. The October 13 changelog says it is GA for Copilot Pro+, Max, Business, and Enterprise, rollout gradual. Copilot lists it as Powerful at $2.00 input, $0.10 cached, and $10.00 output per million until December 1, 2026. Sources: October 13 changelog (opens in a new tab), models and pricing (opens in a new tab).
No letter moves. The Claude 5.5 family and GPT-6 stay in A on the Frontier Models board. S stays locked. AI Coding Agents stays put. Harness rows: tools.
What to do today
- Open the model picker in Copilot Chat, inline edits, ask mode, agent mode, and completions. Note any pin on the three retiring models.
- Point Opus pins at Claude Opus 5.5. Point Terra pins at GPT-6.1 Sol for heavier coding work, or GPT-6 Luna for token-billed volume. That split follows the category labels and prices above. We have not compared quality.
- Search the repo and workflow configs for the old names. GitHub says to update workflows and integrations before November 16. The post does not name Copilot CLI or code review.
- Business and Enterprise admins: if the global default is off, or the replacement was explicitly disabled, enable its model policy. Otherwise GitHub says it is already on. No cleanup task is required after the old models leave.
- Annual-plan individual subscribers: confirm Opus 4.8 is still the Opus in your picker. Do not assume Opus 5.5 appears. Opus 5 and GPT-5.6 Terra are outside this exception.
The weekly patch is on beehiiv: subscribe free (opens in a new tab).
Sources
Checked October 20, 2026.
- Upcoming deprecation, October 15, 2026 (opens in a new tab)
- Models and pricing (opens in a new tab)
- Model multipliers for annual plans (opens in a new tab)
- Supported AI models (opens in a new tab)
- Selected models deprecated, August 31, 2026 (opens in a new tab)
- Claude Opus 5 in Copilot, July 24, 2026 (opens in a new tab)
- Gemini 4 Argon in Copilot, October 13, 2026 (opens in a new tab)