MAI Code 2 is Microsoft’s new coding model, and it’s now showing up in GitHub Copilot at a fraction of what rival models charge. On Microsoft’s own benchmarks it lands close to GPT-6.1 Sol and Claude Opus 5.5. Nobody outside Microsoft has reproduced those numbers yet.
Here’s the short version before the details:
- Microsoft announced it on November 17, 2026, at Ignite in San Francisco, per Reuters.
- There are two models: MAI Code 2 and a cheaper, lighter MAI Code 2 Flash.
- MAI Code 2 is in GitHub Copilot on paid plans, and Flash also reaches Free and Student users.
- Promo prices are half of list price until May 17, 2027.
- Most benchmark results come from Microsoft’s internal testing platform.
Those five facts decide most of whether this matters to you. The rest of this post covers the price after the promo, the scores, and the catches.
What MAI Code 2 is, and how it differs from Flash
Microsoft AI describes MAI Code 2 as a general-purpose coding model for everyday enterprise work. GitHub says it targets deep debugging, multi-file changes, refactoring and repository-wide reasoning, according to WindowsForum’s write-up of the announcement.
MAI Code 2 Flash is the smaller sibling, built for quick, cheap tasks like explaining a function or writing a small helper. It replaces the older MAI-Code-1.1-Flash, which Microsoft is deprecating on December 18, 2026.
Microsoft AI chief Mustafa Suleyman framed the whole launch as a pricing strategy. He called it “a hill-climbing machine” that “co-optimizes performance and price,” as quoted by Reuters. In plain terms, Microsoft wants to win on cost rather than on the top score.
That fits a pattern you may have seen elsewhere. Google pushed the same idea with its coding agents in Antigravity Swarm, and OpenAI keeps trimming prices, as we covered in the GPT-6.2 Astra and Sol pricing post.
MAI Code 2 pricing: the promo and the real list price
Prices below are per million tokens, input then output, as listed for GitHub Copilot by Fello AI. The rival figures are from the same source.
| Model | Promo price (until May 17, 2027) | List price (from May 17, 2027) |
|---|---|---|
| MAI Code 2 | $0.50 / $2.50 | $1 / $5 |
| MAI Code 2 Flash | $0.05 / $0.25 | $0.10 / $0.50 |
| GPT-6.1 Sol | $2 / $10 | $2 / $10 |
| Claude Opus 5.5 | $4 / $20 | $4 / $20 |
The headline claim is that MAI Code 2 costs about 40% of GPT-6.1 Sol. That math uses the promo price. At list price, Fello AI calculates it works out to roughly 81% of GPT-6.1 Sol per task and about 10% of Claude Opus 5.5.
So the discount over OpenAI shrinks a lot after May 2027, while the gap against Claude stays huge. If you compare vendors, use the list column.
How MAI Code 2 scores against GPT-6.1 Sol and Claude Opus 5.5
Microsoft averaged four coding benchmarks: DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1. Here’s how the full model stacks up on its figures.
| Model | Average score | Cost per task (promo) |
|---|---|---|
| MAI Code 2 | 63.2% | $0.309 |
| GPT-6.1 Sol | 64.1% | $0.761 |
| Claude Opus 5.5 | 65.3% | $6.306 |
A point or two behind on average, for a fraction of the cost per task. That’s the pitch, and on paper it holds up.
The weak spot is SWE-Atlas, where MAI Code 2 scores 43.1% against 57.5% for Opus 5.5. Microsoft also says both models are “not yet tuned” for harder problems and very long, autonomous agent runs.
MAI Code 2 Flash looks stronger relative to its price. It averages 61.0% on three benchmarks, versus 55.1% for GPT-6 Luna and 53.9% for Claude Haiku 5.5, according to Fello AI.
Two caveats matter. Microsoft ran most tests on its own MSBench platform, and only FrontierCode came from an outside lab, Artificial Analysis. Its charts also compare against GPT-6.1 Sol, not GPT-6.2, which launched the same day.
Where you can use MAI Code 2 today
Access runs mostly through GitHub Copilot. Based on GitHub’s announcement as summarized by WindowsForum and Apiyi, here’s who gets what.
| Where | MAI Code 2 | MAI Code 2 Flash |
|---|---|---|
| Copilot Free and Student | No | Yes, via Auto only |
| Copilot Pro, Pro+, Max | Yes | Yes |
| Copilot Business and Enterprise | Yes, admin must enable | Yes, admin must enable |
| Microsoft Foundry | Not confirmed | Limited-access preview |
| OpenRouter, Vercel, Kilo and others | Coming soon | Coming soon |
If your company runs Copilot Business or Enterprise, nothing will appear in the model picker until an admin turns on the model policy. Apiyi also notes that Free and Student users can’t pick Flash by hand, because it only appears through Auto.
Rollout reports differ slightly. GitHub’s changelog reads like a live release, but one outlet reported no rollout date, so don’t be surprised if it hasn’t reached your account yet. Apiyi lists a context window of roughly 1.05 million tokens, though real product limits may be lower.
Frequently Asked Questions
Is MAI Code 2 free?
Not the full model. MAI Code 2 needs a paid Copilot plan, while MAI Code 2 Flash is also open to Free and Student users through Auto.
Is MAI Code 2 better than Claude Opus 5.5?
Not on Microsoft’s own numbers. It averages 63.2% against 65.3% for Opus 5.5, and trails badly on SWE-Atlas. It’s much cheaper per task, though.
When does the MAI Code 2 promo price end?
On May 17, 2027. After that, list prices apply: $1 input and $5 output per million tokens for MAI Code 2.
Can I use MAI Code 2 outside GitHub Copilot?
Only in limited ways for now. MAI Code 2 Flash is in a limited Foundry preview that needs Azure approval, and other platforms are marked as coming soon.
Our verdict: try Flash first, keep your hardest work elsewhere
Use MAI Code 2 Flash for routine edits, tests and explanations, and compare it against your current default on a week of real tasks. Keep MAI Code 2 away from your gnarliest, longest-running agent jobs until independent testing backs up Microsoft’s charts. The price is attractive, but a cheap model that needs three retries stops being cheap.


Leave a Reply