Microsoft concedes second place on image quality, then attacks on price: "best price-per-Elo in the world"

Microsoft AI has opened its new image model MAI-Image-2.6 and its fast variant MAI-Image-2.6-Flash to developers. The announcement, dated 4 September, carries its intent in the title: "Pushing the quality-cost frontier."

CEO Mustafa Suleyman shared it from his own account: "Our new image model generates images 2x faster than GPT-Image-2, currently the best model in the world. It is also 72% more efficient in GPU usage, which is why we can offer it at an incredible price. That gives it the best price-performance score in the world."

Concede first, then attack

The opening clause is more striking than the closing claim. The CEO of Microsoft AI calls a rival's model "currently the best model in the world." The company's own blog says the same thing in numbers: MAI-Image-2.6 ranks No. 2 on Arena for both text to image and image editing (as of 4 September). On Artificial Analysis it is No. 2 for text to image and No. 1 for image editing.

So there is no "we are better" claim here. The message is: theirs may be better, but it is not worth the money.

The fine print: which tier?

The speed claim is written differently in the tweet and on the blog, and the difference matters. Suleyman says "2x faster than GPT-Image-2." The blog says MAI-Image-2.6-Flash generates images "2.8x faster than GPT-Image-2-Medium" while delivering 72% greater efficiency.

The difference is the tier. The blog compares against the medium tier, while the quality-price chart Suleyman posted marks the rival as GPT Image 2 (high), the top tier. In the same announcement, speed is measured against the middle tier and price against the top one.

One more separation is worth keeping: the Arena and Artificial Analysis rankings are third party measurements, but the speed and GPU efficiency figures are Microsoft's own.

What OpenAI says

OpenAI has issued no official response to the claim; the latest entries in its newsroom are dated 3 September. But its own pricing page counts as an answer. OpenAI describes GPT-Image-2 as "the most advanced image generation model" and gives a single price: $30 per 1 million tokens for image output, $8 for input, $2 for cached input. Nothing on the page argues speed or cheapness; what it defends is quality.

The same page carries an item that complicates Microsoft's price claim: a 50% discount for batch processing. For a team running production batches, OpenAI's effective price halves. The announcement does not say which price the "best price-per-Elo in the world" claim was built on.

Not a one off, a pattern

Microsoft AI's recent announcement titles all point the same way. MAI-Transcribe-2 is "the fastest, most accurate and cheapest speech recognition model in the world"; MAI-Code-1.1-Flash is "better, faster, at a quarter of the cost". The company is not trying to beat its rival on capability. It is undercutting it on price.

Who is saying it belongs in the picture too. Mustafa Suleyman is a cofounder of DeepMind; after leaving Google he founded Inflection AI, and in 2024 he took over Microsoft AI. Microsoft is OpenAI's largest investor and long standing partner.

What it means for a corporate buyer

For a company running image generation in production, the question is not "which is best" but "how much quality per dollar". Microsoft has made exactly that question askable, which is what lifts this above a product launch. Two things to check when comparing: which quality tier is being measured, and whether the batch discount is in the sum.

Both models are in MAI Playground and in public preview through Microsoft Foundry. The new release adds multi-image reference editing, web grounding, dynamic aspect ratios and output up to 1.5K resolution.