GitHub Copilot Pricing Just Dropped: MAI-Code-1.1-Flash Costs 73% Less

On August 11, 2026, Microsoft made the cheapest serious coding lane in GitHub Copilot a whole lot cheaper. Its new small-tier model, MAI-Code-1.1-Flash, rolled out across GitHub Copilot with a list price 73% lower than the MAI-Code-1-Flash it replaces. For annual Copilot subscribers, it is billed at just a 0.25x premium request multiplier under usage-based billing. For the heavy AI users paying $300 or more a month across tools, this is the rare price cut that actually changes your quota math instead of just your marketing slide.

Microsoft is pricing MAI-Code-1.1-Flash to compete with the Chinese coding models that have been undercutting every Western flagship on price. The result matters more than the headline number, because 73% off a small model does not matter the same way 73% off a flagship does. You need to know what this model is for, how the multiplier works, and where it fits in your Copilot budget before you re-route your agentic workflows to save pennies.

GitHub Copilot Pricing: The New 0.25x Request Multiplier

MAI-Code-1.1-Flash is the successor to MAI-Code-1-Flash, Microsoft’s small-tier coding model. It keeps the lightweight positioning and adds three upgrades: native vision support for reading screenshots and images, better instruction following, and improved tool use so it can drive agents and integrated workflows more reliably. On top of those, continued optimization in model and serving efficiency is exactly what paid for the price drop.

The pricing detail that matters most for heavy users is the request multiplier. Under GitHub Copilot’s usage-based billing, each model is charged at a multiple of a baseline request price. MAI-Code-1.1-Flash is billed at 0.25x for annual Copilot subscribers, and at provider list price for monthly and usage-only billing. A 0.25x multiplier means four requests cost the same as one baseline request, which is the cheapest coding lane in the current Copilot lineup.

To be clear, github copilot pricing has been moving toward usage-based billing for a while. Microsoft deprecated MAI-Code-1-Flash the same day, and the new model is explicitly positioned as the low-cost workhorse for lightweight coding workflows. If most of your agent output is boilerplate, refactors, tests, and straightforward CRUD, this multiplier is where you want it to run.

MAI-Code-1.1-Flash: What the 73% Cut Buys You

A 73% lower list price is the headline, but the multiplier is the number that hits your bill. Here is how the two layers stack for heavy users.

Billing model	How MAI-Code-1.1-Flash is charged
Annual Copilot subscriber	0.25x premium request multiplier
Monthly / usage-based	Provider list price, 73% below MAI-Code-1-Flash
Copilot Free / Student	Included via auto model selection

The split matters. Annual subscribers get the fixed 0.25x discount baked in, which makes per-request cost predictable and easy to budget. Monthly subscribers pay provider list price, which still lands far below the model it replaces but varies with Microsoft’s external pricing. For a team that burns through six figures of coding requests a year, moving the boilerplate-heavy work onto the 0.25x lane can free up a meaningful slice of the monthly bill for flagship work on Claude or GPT-class models.

What has not changed is the ceiling. Even at 0.25x, Copilot requests still draw down your plan’s request pool, and heavy agentic loops still consume quota fast. This is a price cut inside your existing quota system, not a quota increase. The unlimited token myth applies here exactly as it does everywhere else: cheaper per request, still finite in total.

Where It Fits in Your Copilot Budget

The practical play for heavy AI users is to segment your coding work by model tier. Not every task needs a flagship. Massive refactors, deep architecture questions, and gnarly debugging deserve the expensive models. Generating test suites, formatting, simple migrations, and repetitive edits do not, and that is precisely the work MAI-Code-1.1-Flash is built to absorb.

Think of it as a pricing waterfall. Route the high-judgment work to the premium models where a mistake costs more than a request. Route the mechanical work to the 0.25x lane and let the 73% cut pay for your overflow. Teams that already track per-request cost token-by-token will see the biggest win, because they can measure exactly how much of their quota is burning on cheap work that just got cheaper.

There is one operational caveat. MAI-Code-1.1-Flash is off by default for Copilot Business and Enterprise admins, so enabling it is a manual step. If you are on a managed plan, turn the policy on in Copilot settings before you expect your agents to route to it, or your cost savings quietly stay unrealized.

Video Vision, Real Agent Upside

The native vision support is the sleeper feature. MAI-Code-1.1-Flash can read screenshots, UI mockups, and error dialogs directly, which makes it far more useful for agentic coding than a text-only small model. In practice that means a cheap model can now take a screenshot of a broken UI, figure out what is wrong, and write the fix without escalating to a flagship.

That is a big deal for agent workloads that burn flagship tokens on vision-heavy tasks. A significant share of agentic token spend goes to image understanding, and rerouting that onto a 0.25x model is one of the largest per-token savings available in the current Copilot lineup. If you run agents that interact with a browser or a desktop app, test MAI-Code-1.1-Flash on the vision path first.

The Takeaway for Heavy AI Users

The 73% price cut on MAI-Code-1.1-Flash is a genuine cost reduction, not a rebranded quota tweak. For teams running lots of mechanical and vision-heavy coding, the 0.25x multiplier is the cheapest lane in GitHub Copilot today, and it is worth re-routing high-volume work into it immediately.

But keep the ceiling in mind. Cheaper requests still consume your plan’s request pool, and the savings only materialize if you actually route the right work to the cheap model and enable it for managed plans. Segment your workload, enable the policy, and measure the per-request cost. The teams that treat github copilot pricing as a routing problem, not a flat bill, will be the ones who actually keep the 73%.

Microsoft just made the low end of coding dramatically cheaper. The question is whether your workflows are set up to take advantage of it, or whether you are still paying flagship prices for boilerplate work.