Don’t Flip On MAI-Code-1.1-Flash From the Changelog Alone — Watch the September 10 Clock

A changelog that says “73% lower list price” plus native vision is catnip for model-picker tourists. Pause long enough to sync two clocks first.

Core takeaway: MAI-Code-1.1-Flash started rolling out in GitHub Copilot on 2026-08-11, but Copilot Business/Enterprise keep the policy off by default, and MAI-Code-1-Flash is removed across Copilot experiences on 2026-09-10. Do not mash the legacy 0.25× multiplier into the same sentence as usage-based $/1M token list prices.

Why it matters

Swapping a lightweight coding model for something cheaper and snappier feels like free performance. This release is less a vibe upgrade and more an ops event.

• The new model (1.1) and the retirement of 1.0 shipped as a paired announcement.
• On org plans, developers may never see 1.1 until an admin flips policy.
• Teams that parked on MAI-Code-1-Flash as the Sep 1 alternative for Raptor Mini now need a second hop before Sep 10.

If agents, IDE defaults, or internal runbooks hard-code MAI-Code-1-Flash, the failure mode on Sep 10 is not “slightly worse answers” — it is “model missing.”

How it compares

From the GitHub Changelog (2026-08-11), the practical delta looks like this:

Microsoft AI’s product post claims +22% on Terminal-Bench 2.1 in Copilot CLI, +15% on .NET tasks, 25% faster streaming, 25% fewer tokens per task, +4% code survival, and +9% return visits. Treat those as vendor claims and re-check against your own golden tasks.

This is also a different calendar from the Sep 1 six-model retirement covered earlier. “We already updated the September spreadsheet” is how you miss the MAI 1.0 hard cut on the 10th.

Config / policy example

Pin a one-page runbook before anyone “just switches the picker.”

# Copilot MAI dual-clock (as of 2026-08-14)

## Clock A — availability
- [ ] Business/Enterprise: enable policy **MAI-Code-1.1-Flash** (off by default)
- [ ] Confirm picker visibility: VS Code / github.com Chat / Copilot CLI
- [ ] Free/Student: confirm auto model selection path (paid SKUs can also select manually)

## Clock B — deprecation
- [ ] Grep hard-coded `MAI-Code-1-Flash` in agents, CI prompts, team docs
- [ ] Cutover target: **2026-09-10** → `MAI-Code-1.1-Flash`
- [ ] If you moved Raptor Mini → MAI-Code-1-Flash for the Sep 1 note, schedule the second hop

## Billing (do not mix)
- [ ] Usage-based: read provider list $/1M tokens for 1.1
- [ ] Legacy annual request-based: 0.25× multiplier applies only in that billing model

Picker changes without config greps are how ghost model IDs survive.

{
  "copilot": {
    "preferredLightweightModel": "MAI-Code-1.1-Flash",
    "retiredAfter": "2026-09-10",
    "doNotUse": ["MAI-Code-1-Flash"]
  }
}

(That JSON is a team convention example, not a GitHub-shipped schema.)

Practical use

1. Admins before individuals. On Business/Enterprise, “I don’t see the model” tickets arrive before any quality debate.
2. Park 1.1 on cheap loops. Refactors, repo Q&A, short agent turns — keep long design arcs on your frontier route until you have local evidence.
3. Evaluate vision separately. Screenshot-to-code is a new axis. Text-only teams will under-read the changelog if they only hunt for discount language.
4. Split the billing sentence. “0.25×” is a legacy annual request multiplier story; usage-based teams should quote the list $/token table. Finance hates fused folklore.

I nearly pasted “everyone move to 1.1” into Slack from the 73% headline, then discovered half the org still had the Business policy off — so only some pickers showed the model. The fixed order became policy screenshot → picker check → agent-config grep.

Senior-engineer perspective

The model dropdown is packaging. The real interface is policy defaults, retirement calendars, and which billing mode your seat is on. Senior work here is less “how smart is 1.1?” and more “what breaks on Sep 10 if we do nothing?”

Also call out the trap for teams that already migrated to MAI-Code-1-Flash for the Sep 1 Raptor Mini alternative: that landing zone was a layover, not a terminal. Put that sentence at the top of the migration doc so nobody marks the ticket done twice.

Using it in Cursor

If your shop runs Cursor and Copilot side by side, split jobs instead of mirroring every model flip.

• Cursor: long-horizon agents, multi-file refactors, harness-heavy repo work (draft APIs, WIP boards, skills)
• Copilot 1.1: lightweight IDE turns, short CLI jobs, screenshot-led UI sketches

Add a Cursor rule/skill note like “team lightweight default on the Copilot side = MAI-Code-1.1-Flash” so chats stop renegotiating the picker. Enabling a model in Cursor does not enable a Copilot org policy — different product, different gate.

FAQ

Q. I don’t see 1.1 in the Business picker.
A. Have an admin enable the MAI-Code-1.1-Flash policy. Changelog: off by default for Business/Enterprise.

Q. Does 0.25× mean every bill is quartered?
A. No. That multiplier is for legacy annual request-based billing. Usage-based seats should read the list $/1M token table.

Q. Is the Sep 1 deprecation enough?
A. No. MAI-Code-1-Flash has a separate Sep 10 retirement.

Q. Can we trust Microsoft’s +22% claim?
A. Log it as a vendor claim; re-measure with your golden CLI / .NET / vision tasks.

Q. Do Free/Student admins need the policy toggle?
A. Changelog describes Free/Student availability via auto model selection. Don’t confuse that path with org policy gates.

Sources

• MAI-Code-1.1-Flash available in GitHub Copilot
• Upcoming deprecation of MAI-Code-1-Flash
• MAI-Code-1.1-Flash: Better, faster, at a quarter of the cost (Microsoft AI)
• Copilot models and pricing
• Model multipliers for annual plans (legacy)
• Upcoming August 2026 model deprecations in GitHub Copilot

Closing

Cheap model patches feel good; policy defaults and retirement dates move the outage risk. This week, deleting old model strings before Sep 10 beats bragging about the picker. The lightweight-model market will likely reward teams who miss fewer migration clocks more than teams who quote the newest bench row.