Microsoft launches new AI models for GitHub Copilot programming, promising cost cuts

2–3 minutes

The software giant unveiled its new MAI Code 2 and MAI Code 2 Flash models at the Ignite conference in San Francisco on Tuesday, 17th. These models will now be integrated into GitHub Copilot, with the aim of reducing the cost of using artificial intelligence for coding and code review, thereby strengthening Microsoft’s position against competitors like OpenAI and Anthropic.

Functions and availability of the new models

Code 2 is designed for complex tasks, such as investigating code errors and implementing changes across multiple files. Flash, on the other hand, is tailored for shorter, more specific tasks, like explaining code snippets or making targeted corrections.

Both models are available to eligible GitHub Copilot users. The primary model is included in paid plans, while Flash is also accessible through free and student plans. In Microsoft Foundry, Flash is currently in limited access preview.

What changes in the cost of the more affordable model?

According to Microsoft, in four programming tests, Code 2 achieved an average of 98% of GPT-6.1 Sol’s score at 40% of the cost, and 96% of Claude Opus 5.5’s score at 5% of the cost.

The comparison uses Code 2’s promotional price, which includes a 50% discount. Three tests were conducted on the company’s internal platform, while the fourth used data from Artificial Analysis.

These percentages reflect results from specific tests and do not represent the total cost of software projects. Microsoft clarifies that the model was optimized for common corporate programming tasks, not for scientific research or long-term freelance work.

A derivative version of Code 2 Flash, called MAI Work Flash, has been adapted for Microsoft 365 applications. Internal tests included tasks in Excel, Word, PowerPoint, and Teams, without excluding other models in these programs.

Other announcements made at Ignite

  • Image: MAI Image 2.7 and 2.7 Flash are now in public preview on Foundry. Microsoft estimates costs of around $35 and $10, respectively, per thousand images generated. Flash also powers the experimental Canvas editor from MAI Labs.
  • Voice: MAI Realtime 1 responds via audio, with pricing starting at $0.01 per minute of audio and separate charges for reasoning processing. A demo is available at MAI Labs, and a limited preview is available on Foundry.
  • Digital security: MAI Cyber 2 Flash will be exclusively available for Microsoft security products starting mid-December for eligible customers.

The company also announced plans to license the weights of its models to select customers. These weights are numerical values adjusted during training and will not be publicly released. This initiative reinforces the strategy announced in October, when Microsoft unveiled the pre-training of MAI-Base-2.

Internal development is ongoing alongside the non-exclusive licensing of OpenAI’s technology, which remains valid until 2032. At the same event, Satya Nadella announced the arrival of GPT-6.2 on Foundry.

Frequently Asked Questions

What is the difference between MAI Code 2 and Flash?

Code 2 was developed for more complex code changes, while Flash was optimized for frequent, specific tasks like explaining code and making small corrections.

Where are the programming models available?

Both models are available on GitHub Copilot for eligible users. Flash is also in limited access preview on Microsoft Foundry.

Is MAI Code 2 always cheaper than Claude?

It’s not possible to conclude this based on the published tests. Microsoft’s comparison uses specific tasks and a promotional price for Code 2.