IA4 MIN

Microsoft's MAI Code 2 lands in GitHub Copilot, claiming 96% of Opus 5.5 at 5% of the cost

It is one of seven in-house models Microsoft unveiled at Ignite. The Flash version even reaches GitHub Copilot's free tier, while the consumer Copilot app is not on the list of places they are headed.

A pair of programming curly braces drawn in glowing text characters floats over a dark background with a band of golden haze
Image: Microsoft AI

Microsoft used its Ignite conference in San Francisco on Nov. 17 to unveil seven AI models built by its in-house MAI (Microsoft AI) division. The one developers will notice first is MAI Code 2, already in GitHub Copilot alongside a faster, cheaper sibling, MAI Code 2 Flash. In Microsoft AI's Ignite roundup, the company says Code 2 delivers 96% of the performance of Anthropic's Claude Opus 5.5 at 5% of the cost.

Those figures come from Microsoft's own tests and math, with a promotional discount applied to its model.

01

What MAI Code 2 does and who gets it

MAI Code 2 is meant for heavier coding work, such as tracking down a stubborn bug or making a change that touches many files in a project. Code 2 Flash takes the quick, repetitive jobs, like explaining a block of code or fixing a routine error.

Code 2 is available on Copilot Pro, Pro+, Max, Business and Enterprise, and Flash also reaches the no-cost Free and Student plans, according to GitHub's changelog. Both appear in the model picker in VS Code, JetBrains, github.com and Copilot's other tools. GitHub says access is for "eligible" subscribers on annual plans, without spelling out who qualifies.

Copilot bills by usage. Each plan comes with AI credits (Pro is $10 a month with 1,500, per GitHub's plans page), but GitHub has not said how many MAI Code 2 consumes. The model also joins the Standard tier of Copilot Fusion, the mode that picks an AI for each task on your behalf.

A presenter in a white sweater on the blue Ignite 2026 stage, below a screen showing icons for Code, Decision, Cyber, Image, Transcribe, Custom Models and Licensed Weights
Image: Microsoft AI
02

Reading the "96% at 5% of the cost" claim

Microsoft averaged four benchmarks that hand a model programming tasks, from fixing bugs to working in a terminal. It ran three of them on its own systems, according to its MAI Code 2 announcement. The fourth comes from Artificial Analysis, an independent firm that compares AI models.

The 5% is not a per-token price, the unit of text these models bill for. It is what it cost to complete those tests, with MAI Code 2 costed at a promotional rate that halves its normal price. At full price, that roughly doubles to about 10% of what the same tests cost with Opus 5.5. Against OpenAI's GPT-6.1 Sol, Microsoft claims 98% of the performance at 40% of the cost, or about 80% without the discount.

Microsoft has not published a per-token price for MAI Code 2, and no outside party has rerun the full comparison yet. The company itself says the model is not tuned for scientific research or for very long jobs in which the AI works through many steps on its own.

03

The other five models

The rest of the lineup is aimed mostly at businesses, inside Microsoft products or through Foundry, the Azure platform where Microsoft sells AI models.

  • MAI Cyber 2 Flash hunts for software vulnerabilities, suggests fixes and investigates threats. It will run only inside Microsoft's security products, starting in mid-December. Output costs $0.50 per million tokens versus $50 for GPT-6 Astra on Azure, and Microsoft reports a 92.3% score on the CyberGym-E2E benchmark against Astra's 91.5%.
  • MAI Image 2.7 creates and edits images. Microsoft says it leads Artificial Analysis's image-editing rankings, and its Flash version costs about $10 per 1,000 images.
  • MAI Realtime 1 is for building voice assistants that listen and talk back live with little lag, starting at one cent per minute of audio.
  • Dragon Copilot Transcribe transcribes doctor visits in Dragon Copilot, Microsoft's clinical assistant. Microsoft says a third of U.S. physicians use it.
  • Decision 1.1 routes requests between models so the expensive ones only run when needed.
Official MAI Cyber 2 Flash artwork showing several overlapping white shield shapes on a blue and violet background
Image: Microsoft AI
04

Licensed Weights: the model inside the customer's cloud

A model's weights are the numbers it tunes during training, and they hold what it has learned. With MAI Licensed Weights, Microsoft will give selected companies those weights so they can run the model in their own Azure environment, adapt it with their data and decide who uses it. Microsoft says it will not train on that data without explicit agreement. The program is "coming soon," with no price or date.

05

What changes for Copilot users

It depends on which Copilot you use. The free consumer app on Windows, phones and the web does not appear in Ignite's model announcements, so there is nothing new to turn on. For Microsoft 365 Copilot, the workplace version that already bundles chat and agents through Home, Code and Autopilot, Microsoft has spun off a Word, Excel, PowerPoint and Teams model from Code 2 Flash, MAI Work Flash, but has not said who has it yet.

If you code with GitHub Copilot, you can try MAI Code 2 Flash on the free plan and MAI Code 2 from Pro up. Microsoft is not sidelining OpenAI, either. At the same event, Satya Nadella announced that GPT-6.2 is coming to Foundry.

00

The conversation starts here

Sign in with a supporter account to comment. Sign in

Nobody has commented yet. Want to go first?

YOUR NEXT ROUTE

Keep following AI models and agents

If this story interests you, these three pieces are the best place to carry on.

OPEN THE FULL TOPIC
  1. 01Claude Sonnet 5.5 gets close to Opus. The real surprise is what it costs to runIA · 4 MIN
  2. 02Claude Opus 5.5 cuts prices and earns promising early reactions, with caveatsIA · 2 MIN
  3. 03Google’s 300-language milestone does not mean 300 offline languagesIA · 2 MIN

KEEP READING

You may also like

FRONT PAGE