GitHub added MAI-Code-1.1-Flash to Copilot and priced it at one quarter of a premium request. The model it replaces disappears on September 10, so this is both a cost change and a migration deadline.
MAI-Code-1.1-Flash is rolling out across GitHub Copilot. Annual Copilot subscribers pay a 0.25× premium-request multiplier when they select it. Free and Student users can receive the model through automatic selection, while paid individual and managed plans can select it manually where administrators allow it.
GitHub says the update improves coding, instruction following, tool use, and performance, adds native image understanding, and cuts the provider’s list price by 73%. Those are launch claims. The useful evidence for a team will come from its own repositories.
The 0.25× multiplier changes the routing calculation
A quarter-request multiplier means four eligible MAI-Code-1.1-Flash requests consume the same premium-request allowance as one 1× request, before plan-specific rules and rounding. That makes it a plausible default for routine coding chat and agent work if its completion quality holds up.
| Copilot user | How MAI-Code-1.1-Flash appears | Billing signal |
|---|---|---|
| Free or Student | Automatic model selection | Plan limits still apply |
| Annual paid individual | Manual or automatic selection | 0.25× premium request |
| Business or Enterprise | Manual or automatic, subject to policy | Managed plan rules apply |
The multiplier is not a quality grade. It is a consumption unit. A cheap request that needs three repair passes can cost more developer time than a pricier model that lands the change once.
Native image input widens the tasks you can test
Image understanding lets developers bring screenshots, diagrams, and visual bug evidence into a Copilot conversation. That could help with UI reconstruction, log-dashboard interpretation, and explaining a failing visual state. Do not assume that an image-aware model understands the hidden application state behind a screenshot. It sees pixels and the context you provide, not the missing network call or database row.
A good evaluation should pair each image with the actual repository context and a checkable task: identify the component, propose the smallest repair, name what cannot be inferred, and produce a test. Score the result on correctness and review burden, not on how confident the explanation sounds.
September 10 is the hard part of the announcement
GitHub will remove the older MAI-Code-1-Flash model from Copilot on September 10, 2026. The recommended replacement is MAI-Code-1.1-Flash. Saved instructions, screenshots, agent tools, and repository rules that worked with the old model should be retested before that date.
- Enable it deliberately. Managed organizations should confirm the model policy before asking developers to migrate.
- Build a small repo test set. Include bug fixes, refactors, tests, tool calls, and one screenshot-based task.
- Measure repairs. Count retries, review comments, reverted changes, and total premium requests.
- Check the fallback. Verify which model automatic selection chooses when the preferred model is unavailable.
- Remove the old name. Update team docs and routing rules before September 10.
Do not confuse a provider price cut with your total cost
GitHub’s 73% figure refers to the provider list price. Copilot users experience model economics through premium-request multipliers, plan allowances, latency, and the human cost of review. Those measurements can point in different directions.
Our article on Copilot review effort and agent usage metrics explains why adoption counters should not become quality scores. The same warning applies here. The Kimi K3 rollout pause in Copilot is another reminder that production behavior belongs beside model claims.
My verdict: route the cheap work, then audit the repairs
MAI-Code-1.1-Flash is attractive because the migration target and the low multiplier arrive together. I would test it first on frequent, reversible tasks where a quarter-request price can compound into real savings.
I would not make it the default for security boundaries, migrations, or unfamiliar systems until the team has measured repair rate and reviewer time. September 10 forces the model change. It does not force you to skip the evaluation.
Read the primary sources
- Read GitHub’s MAI-Code-1.1-Flash availability note.
- Review GitHub’s September 10 deprecation notice.
Checked August 12, 2026. Capabilities, price comparison, multiplier, plan availability, and retirement date are reported by GitHub.
