Microsoft launches seven MAI models at Ignite, pitching its coding model at 5% of Claude Opus's price

AI & Compute

Microsoft’s seven MAI models put a 5% coding-price claim center stage

Microsoft unveiled seven MAI models on Nov. 17. MAI Code 2 claims 96% of Anthropic's Claude Opus 5.5 coding performance at about 5% of the price, without independent validation.

By
Sophie Lindqvist
Filed
Channel
AI & Compute
Read
3 min read

Sophie Lindqvist

Microsoft announced seven in-house MAI models at its Ignite conference on Nov. 17, claiming MAI Code 2 reaches 96% of Anthropic's Claude Opus 5.5 coding performance at about 5% of its price.

Microsoft supplied most benchmark and price comparisons. The company did not publish independent benchmark results.

What does Microsoft claim for coding?

Microsoft says MAI Code 2 averaged its reported results across DeepSWE 1.1, Terminal Bench 2.1, SWE-Atlas and FrontierCode, reaching about 98% of OpenAI's GPT-6.1 Sol at roughly 40% of its price.

In a separate blind front-end test, developers preferred MAI Code 2 over GPT-6 Sol 64% of the time, Microsoft said.

MAI Code 2 is live in paid GitHub Copilot plans.

Microsoft prices MAI Code 2 Flash at 20% of Claude Haiku 5.5 and says the model matches GPT-6 Luna at 65% of its price. The Flash model reaches Copilot's free and student plans and has a limited preview on Microsoft Foundry.

Microsoft gave no dollar or per-token price for either coding model.

What does the cyber model cost?

Microsoft says MAI Cyber 2 Flash scored 92.3% on the CyberGym end-to-end benchmark, compared with 91.5% for OpenAI's GPT-6 Astra, while costing under 2% per task.

The model has a 1-million-token context window. Microsoft lists $0.10 per million input tokens and $0.50 per million output tokens.

Cyber 2 Flash remains limited to Microsoft's security products. It will replace the current model in the MDASH security service for eligible customers from mid-December.

Where does image editing stand?

Microsoft prices MAI Image 2.7 at about one-fifth of GPT Image 2.6 for image editing, while Artificial Analysis ranks it first for that task, ahead of OpenAI's model, and fourth for text-to-image generation.

MAI Image 2.7 is in public preview on Microsoft Foundry and through fal, OpenRouter and Vercel.

Which models are available now?

Microsoft's seven-model lineup does not share one launch status.

  • MAI Realtime 1 is Microsoft's first live speech-to-speech model. Pricing starts at $0.01 per minute.
  • Dragon Copilot Transcribe, built on MAI Transcribe 2, handles about 10 million patient encounters each month, although Microsoft did not disclose a price for that service.
  • Decision 1.1 provides model routing at $0.042 per million input tokens.

Microsoft calls the release seven models, yet the announcement also describes a high-volume MAI Image 2.7 Flash. It does not explain how that variant fits the count.

What does licensed weights change?

Under the program, an enterprise can take a frontier MAI model, customize it with its own data, and run it inside its own Azure environment while retaining control of access.

Microsoft said in the announcement: “No one wants to pour their IQ into someone else's model.”

Microsoft also said its second-generation Maia 200 AI accelerator is now in production, running its largest models. The company disclosed no process node, wafer size, production capacity or shipment figure for Maia 200.

That establishes production status, not commercial scale. No accelerator revenue figure was provided.

What remains unproven?

The launch also lands against evidence of uneven Copilot adoption: Microsoft's own data shows Anthropic's Claude has less than one-tenth of Copilot's weekly users inside Office, The Information reported earlier Tuesday.

That ratio does not reveal absolute usage or revenue. For MAI Code 2 and MAI Image 2.7, Microsoft gives relative rather than absolute prices.

Most performance comparisons still come from Microsoft. Artificial Analysis provides a separate ranking, not validation of every benchmark score.

From mid-December, eligible MDASH customers will receive Cyber 2 Flash, making security-product deployment a near-term test of Microsoft's claimed sub-2% task economics against GPT-6 Astra.

Source: mlq.ai

Share this article:

More from Sophie Lindqvist

Sophie Lindqvist

Show full bio

News editor covering money, funding and market shifts around Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,113 articles

Related articles

« Previous articleNext article »