Seven models, each framed as a discount
Microsoft’s AI division announced seven new MAI models on 17 November at its Ignite conference, and the striking thing about the announcement is its structure. Almost every model is introduced not by what it can do, but by how close it comes to a named competitor and at what fraction of the price.
MAI Code 2, Microsoft says, reaches 98% of the coding quality of OpenAI’s GPT-6.1 Sol at about 40% of the price, and 96% of Anthropic’s Claude Opus 5.5 for 5% of the price. MAI Code 2 Flash matches GPT-6 Luna at 65% of the price and Claude Haiku 5.5 at 20%. MAI Cyber 2 Flash, Microsoft says, matches GPT-6 Astra’s quality on the CyberGym-E2E benchmark at less than 2% of the cost per task.
The rest of the lineup
MAI Image 2.7 is described as first on Artificial Analysis’s image-editing ranking, beating GPT Image 2.6 Sunburst at maximum settings for a fifth of the price. MAI Realtime 1, a voice model, is said to outperform GPT-Realtime 2.1 and starts at $0.01 per audio minute. Decision 1.1, the newest of Microsoft’s decision models, is described as 50 times faster and 79 times cheaper than GPT-Sol 6.1.

The seventh entry is a product rather than a bare model: Dragon Copilot Transcribe, built on MAI Transcribe 2, which Microsoft calls the world’s fastest and most accurate multilingual transcription model. The company says it already powers 10 million patient encounters a month and has halved multilingual error rates.
Every one of these figures is Microsoft’s own, measured on its own configurations. None has been reproduced by an independent evaluator.
Why a reseller is undercutting its supplier
Microsoft is OpenAI’s largest commercial partner and sells OpenAI models through Foundry. It is now also selling models that it positions explicitly as cheaper substitutes for them. The stated reason, in Microsoft’s own words, is that token spend is rising and customers want prices that let them run agents around the clock doing real engineering work.
That argument is the same one OpenAI made on the same day when it cut GPT-6.2 Astra’s token prices by 20% and emphasised cost per finished task over raw capability. Two of the largest vendors in the market spent 17 November competing on price.

Maia 200 is in production
Alongside the models, Microsoft said its second-generation AI accelerator, Maia 200, is now in production and running the company’s largest models, achieving 1.7 times more tokens per watt. Microsoft first announced the chip in January 2026 and said at the time that it would say more about its strategy at Ignite.
The chip matters to the pricing claims. Running its own models on its own silicon is the mechanism by which Microsoft can plausibly charge a twentieth of what a rival charges for comparable output, and it is the part of the announcement that a competitor cannot answer with a price cut.
What is not settled
Microsoft has not published independent verification for any of the comparisons, and several of the named rivals — GPT-6 Luna, GPT-Sol 6.1, GPT Image 2.6 Sunburst — are compared at specific settings that the announcement does not fully describe. The CyberGym-E2E claim comes with its own caveat disclosed on Microsoft’s separate page for that model, where the company says it excluded 92 of the benchmark’s 905 tasks because they overlapped its training data.
What to watch is whether Artificial Analysis and other outside evaluators publish their own numbers for MAI Code 2 and MAI Image 2.7 in the coming weeks, and whether MAI Licensed Weights, announced the same day, turns the price argument into deployments inside customers’ own Azure tenancies.