Language and display
BE THE BOSS OF AI.ADRIAN DUNKLEY ↗
← This newspaper edition

Check what could change when GitHub retires your Copilot model

GitHub will drop several versions from its coding assistant on 16 November. Save a small test before moving a job to a replacement.

A blue arch and a lavender arch face each other across a gap, with a red cut-out in front.
Spare Piece at the Crossing · Editorial artwork · AI Boss Ltd

Find any job your team set up to use a particular model in GitHub Copilot, an AI assistant for writing and checking code. GitHub says it will retire Claude Opus 4.8, Claude Opus 5 and GPT-5.6 Terra from Copilot on 16 November. It will keep Opus 4.8 available to people on individual annual Copilot plans. The dates apply to versions available through Copilot.

Check which options your team can open

GitHub suggests Claude Opus 5.5 in place of the Opus versions and offers GPT-6.1 Sol or GPT-6 Luna as alternatives to GPT-5.6 Terra. It says these choices are on by default for Copilot Business and Enterprise unless an administrator switched them off. Its support table also shows that the models you can choose depend on your plan and which app you use. Ask the person who maintains the job to look for the old model’s full name in saved settings. Have the person who runs the job check the model menu in their own account before you choose a replacement.

The 13 November changelog lists another round for 14 December, including Claude Fable 5 and Gemini 3.8 Flash. GitHub names Fable 5.2 and Gemini 4 Hydrogen as their respective alternatives. Its notice leaves the replacement for Kimi K3 open. If a team uses several models, record the name, job and backup for each when you next review the list.

See whether the replacement catches a missing order

Try a made-up order reconciliation before entrusting a model with a live shop report. Use a new Copilot Chat session for each model in the same app. Paste this instruction into each session: “Compare these lists by order number. List missing and changed orders. Ignore the row order. On the register: A12=3, B13=2, C14=1. On the packing list: B13=2, A12=4. Use only those records.” Leave each model’s settings at their defaults and record the names and results. The result should show that C14 is missing from the packing list and that A12 has one extra unit packed. B13 should stay off the list because both records give the same count. If the old model has gone, compare the candidate’s output with those expected results.

Repeat the check on an approved sample of the job you want to move. Add examples your team has had to fix before. Keep the existing human review in place and move the job only after the replacement handles each of those cases.

Weigh a move away from your supplier

For a firm weighing whether to keep using a vendor or operate some of the technology itself, open the Sovereign AI Cost-Benefit Analysis for Companies. Select Start the Analysis and answer for the workload you want to move. The tool asks about the sensitivity of the data, work volume and cost, staff and infrastructure, and how much your business depends on AI. You can select View Results Without Sharing Contact Details after the questions.

A twelve-question practice run returned “Use the APIs” for a company that reports a concern about depending on one supplier. In the exercise, the company mainly used public information, spent under US$500 a month on AI, had one or two technology generalists and made limited use of AI in its services. An API lets one application call another. The recommendation comes from the company’s answers, including its reported spending and staffing. Calculate the cost of moving the job from actual invoices and staff time before deciding who should run it. Keep the order-comparison instructions, approved test records and correct results where your team can use them with another service. Ask someone who did not write the instructions to use those records to run the comparison.

Sources and further reading

Sources checked 15 November 2026.