ModelBrief · MAI Code 2 · Topic Page
Last updated Nov 20, 2026

MAI Code 2 Pricing: $0.50/$2.50 Promo, Doubles May 2027

ModelBrief · Published 2026-11-20 · Fact-check 2026-11-20 · Reference: Microsoft AI product post, GitHub changelog and pricing page, Axios report

MAI Code 2's pricing story has two layers: Microsoft's product post gives only relative costs, but GitHub's Copilot pricing page already publishes the rate card — $0.50/$2.50 per 1M tokens on a promotional price that ends May 17, 2027 and doubles after. This article lists the rate card first, then takes apart the relative-cost figures Microsoft did give: how they were computed, what they hide, what may change when the promotion ends, and how you actually pay inside Copilot.

Only four things are confirmed

ItemConfirmed
Rate cardGitHub's Copilot pricing page publishes it: MAI Code 2 $0.50 input, $0.025 cached input, $2.50 output per 1M tokens (promo to May 17, 2027), then $1.00/$0.05/$5.00; Flash $0.05/$0.005/$0.25, then $0.10/$0.01/$0.50
Relative costsMAI Code 2: about 40% of GPT-6.1 Sol, about 5% of Claude Opus 5.5; Flash: 65% of GPT-6 Luna, 20% of Claude Haiku 5.5 (all Microsoft-reported)
Promotional priceBoth models shown at a 50% promotional price (Microsoft product-post footnote, GitHub pricing page); the promotion ends May 17, 2027, and prices double after
BillingUsage-based inside GitHub Copilot (GitHub changelog); limited to eligible subscribers on annual plans

How "40%" and "5%" were computed

Microsoft's cost comparison rests on two layers of assumptions. The first is the benchmark basis: costs are compared as average cost per task across four benchmarks (DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas, FrontierCode v1.1), all models at max reasoning effort. The second is the price basis: MAI Code 2's cost is computed at the 50% promotional price, while rivals' costs use their list prices. Microsoft's official blog chart gives the average cost per task: MAI Code 2 at $0.309, GPT-6.1 Sol at $0.761, Claude Opus 5.5 at $6.306 — $0.309 ÷ $0.761 ≈ 40%, and $0.309 ÷ $6.306 ≈ 5%. That is where "about 40%" and "about 5%" come from. Note the asymmetry: the rivals are not discounted, MAI Code 2 is half-price. The comparison is asymmetric by construction — a fact stated in Microsoft's own footnote, not this site's inference.

After the promotion: 40% becomes 80%, 5% becomes 10%

Working backwards from Microsoft's own footnote, the promotional price is half the list price. After the promotion, MAI Code 2's average cost per task rises from $0.309 to about $0.618 (assuming usage and model behavior unchanged), its relative cost vs GPT-6.1 Sol from about 40% to about 80%, and vs Claude Opus 5.5 from about 5% to about 10%. Flash follows the same arithmetic: vs GPT-6 Luna from 65% to about 130% — meaning after the promotion Flash would be more expensive than Luna; vs Claude Haiku 5.5 from 20% to about 40%. The "about 130%" figure deserves its own sentence: it means Flash's price advantage over Luna depends entirely on the promotion and reverses the moment it ends. The promo end date is now published (May 17, 2027, GitHub pricing page); the post-promo relative costs remain this site's arithmetic worked backwards from Microsoft's footnote, not figures Microsoft published — though the about-$0.62 per-task figure matches Microsoft's own numbers as reported by multiple outlets.

Copilot usage billing: how you actually pay

MAI Code 2 has no standalone API; the bill runs through GitHub Copilot's usage-based billing. GitHub's changelog confirms both models are usage-billed and limited to eligible Copilot subscribers on annual plans; under Business and Enterprise plans, admins must allow MAI models in the Copilot model policy (enabled by default). That means three things: first, your bill depends on usage, not a fixed price list on top of a flat monthly fee; second, actual charges follow GitHub's Copilot pricing page: MAI Code 2 at $0.50/$2.50 (promo to May 17, 2027), then $1.00/$5.00; third, the promotional price shows up in Copilot billing, so the same usage and the same tasks cost more after the promotion ends — enterprises should stress-test budgets against the post-promo relative costs.

Why AT&T's "50x" cannot be used as pricing

Microsoft says a custom AT&T version of MAI Code 2 Flash outperformed GPT-6.1 Sol on standard 5G network analysis while costing 50x less. The line is often cited as pricing evidence, but it is not pricing: a custom build, one customer, one task type, no published test method, and no statement of which metric "outperformed" refers to. Converting it into general pricing means treating one customer's case as a rate card. This site records it only as a "single-customer anecdote" and uses it in no price comparison.

Axios' warning: the pricing advantage reflects promotional discounts

Axios' launch-day report repeats Microsoft's pricing claims while stating plainly that the pricing advantages reflect promotional discounts. It is the only confirmed tier-1 English media report so far, and it is also this article's core judgment: MAI Code 2's pricing story is not "cheap" but "cheap during the promotion". The promo end date is now published — May 17, 2027 — so the advantage lasts about six months, and the bill doubles when it ends.

Buying advice: budget on the post-promo bill

If you set MAI Code 2 as Copilot's default model now, the short-term bill does fall; but do two things at the same time: first, check the actual charges on GitHub's Copilot pricing page rather than trusting only the relative figures in Microsoft's launch post; second, re-run the budget against the post-promo relative costs (about 80% vs GPT-6.1 Sol, about 10% vs Opus 5.5) — if it still pays off at those figures, then make it the long-term default. Flash users should be more careful still: after the promotion it may cost more than GPT-6 Luna, so compare again then.

ModelBrief's take: MAI Code 2's pricing story is one sentence long: there is a rate card, but it is a promotional price — $0.50/$2.50 until May 17, 2027, then it doubles. The 40% and 5% are real Microsoft figures, but they are relative values inside the promotion, not long-term prices. Budget on the post-promo bill, and the promotion's end won't catch you by surprise.

Reference: Microsoft AI — Product post "MAI Code 2 and Code 2 Flash: Everyday coding at scalable prices" · GitHub Changelog — MAI Code 2 and MAI Code 2 Flash in GitHub Copilot · Axios report (via superpowerdaily) (rewritten and organized by ModelBrief; facts per official sources; the rate card is from GitHub's Copilot pricing page, checked by multiple outlets on Nov 17–18; post-promo relative costs are this site's arithmetic worked backwards from Microsoft's footnote, not Microsoft-published figures)