Microsoft AI announced seven new MAI models on Nov. 17, according to its Ignite post. The coding pair, MAI Code 2 and MAI Code 2 Flash, is already in GitHub Copilot. Microsoft's performance and price comparisons below are its own.

The seven models

  • MAI Code 2: for everyday enterprise coding such as debugging and multi-file changes.
  • MAI Code 2 Flash: for high-volume work such as code explanations and routine fixes.
  • MAI Cyber 2 Flash: a security model. Microsoft says it matches GPT-6 Astra's quality on the CyberGymE2E benchmark at under 2% of the cost per task.
  • MAI Image 2.7: an image model. Microsoft says it ranks #1 for image editing on Artificial Analysis.
  • MAI Realtime 1: a speech-to-speech model, priced from $0.01 per audio minute.
  • MAI Transcribe 2: a speech-to-text model, behind Dragon Copilot Transcribe.
  • Decision 1.1: a model-routing model. Microsoft says it is 50x faster and 79x cheaper than GPT-6.1 Sol.

Microsoft also said its Maia 200 chip is now in production. That is hardware, not one of the seven.

What Microsoft claims for MAI Code 2

Microsoft says it averaged four coding benchmarks: DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode v1.1.

On that average, Microsoft says MAI Code 2 delivers 98% of GPT-6.1 Sol's performance at 40% of the cost, and 96% of Claude Opus 5.5's at 5%. The fine print says all models ran at max reasoning effort and that MAI Code 2's cost is shown at its 50% promotional price.

Microsoft says Code 2 Flash performs comparably to GPT-6 Luna and Claude Haiku 5.5 at 65% and 20% of their cost. Microsoft says both coding models are not yet tuned for harder work, such as science research or very long agent runs.

Where to try it

Per Microsoft's MAI Code 2 post, MAI Code 2 and Code 2 Flash are in GitHub Copilot, and Code 2 Flash is in limited-access preview in Microsoft Foundry. Microsoft says the coding models are coming soon to OpenRouter, Vercel, Hermes, Open Code and Kilo.

MAI Image 2.7 and its Flash version are in Foundry. MAI Realtime 1 can be tried in MAI Labs, with a sign-up for interest. Microsoft also announced MAI Licensed Weights, which lets a customer run a customized MAI model in its own Azure environment.

Who is affected and what this means for you

If you use GitHub Copilot, MAI Code 2 is now an option there. The 98% and 40% figures come from Microsoft's benchmark runs, and the cost side uses a promotional price, so treat them as a starting point and test on your own code.

For other Microsoft AI news, see Microsoft's Copilot deal with Salesforce. Microsoft's comparison models include GPT-6.1 Sol, which OpenAI has since followed with GPT-6.2.

What we do not know

  • Prices: the posts we read give relative costs, not per-token prices, and we do not know when the 50% promotional price ends.
  • Independent results: we have not seen outside tests of these numbers, and we have not run the models.
  • Counts: Microsoft's two posts differ slightly on one result. The Code 2 post says reviewers preferred its frontend work over GPT-6 Sol 67% of the time, and the Ignite post says 68%.