Microsoft’s MAI Image 2.7 Ranks Above Google, Behind OpenAI

On Arena’s image-edit board on 28 October, the only two models above Microsoft AI’s new image-editing model, MAI Image 2.7, were from OpenAI. Microsoft AI said on 27 October 2026 that the model “launches in preview at No. 3 on Arena Edit, making Microsoft AI the No. 2 lab, and No. 2 on Artificial Analysis’s image editing leaderboard.”

Mustafa Suleyman, who leads Microsoft AI, wrote on X that MAI-Image is “once again beating Google, Meta, xAI, and GPT-Image-2.5-Flare.” The model is for editing images for now, and Microsoft says text-to-image generation is “coming soon.”

Short explainer · Microsoft MAI brief

A vertical news Short narrated from the published headlines and opening lines of this piece and four related FourWeekMBA pieces on Microsoft AI and image models.

Business Pill · Benchmarks explained

An evergreen Business Engineer explainer on why a test score is not your own result. It explains the general idea only and does not discuss Microsoft or Arena’s rankings.

Where MAI Image 2.7 Sits on Arena

Arena ranks image-editing models from people’s votes between pairs of outputs. On its image-edit board on 28 October, OpenAI’s GPT-Image-2.5 Sunburst led with a score of 1517 and GPT-Image-2.5 Flare was second at 1482. Arena marks both scores as preliminary.

MAI-Image-2.7-Preview was third at 1471, with a margin of plus or minus 10 points from 3,402 votes, and Arena gives its rank range as 2 to 4. OpenAI’s GPT-Image-2 (Medium) was fourth at 1460 and xAI’s Grok Imagine Image 2.0 fifth at 1439.

Microsoft’s previous model, MAI-Image-2.6, sat sixth at 1427, level with Google’s Gemini Nano Banana 2.1, the highest Google entry on the board. By our arithmetic, MAI Image 2.7 scores 46 points below Sunburst, 11 below Flare and 44 above both MAI-Image-2.6 and Nano Banana 2.1.

What Was Posted on X

Mustafa Suleyman wrote on X on 27 October that MAI-Image is now the No. 2 image edit model in the world, once again beating Google, Meta, xAI and GPT-Image-2.5-Flare, and pointed readers to MAI Playground.

What Microsoft Says Changed

Microsoft AI says that “MAI-Image-2.7 improves on 2.6 across all 10 use cases and all 7 editing actions evaluated by Artificial Analysis.” It says the model reaches the category frontier in enhancement and restoration, reasoning-based edits, scene and style, and text or symbol editing.

The company says the model’s “strongest lead is in scene and style, where it scores 24 Elo ahead of the next-best model.” It names restyling or relighting a scene, changing text inside an image, and restoring existing assets as the edits it targets.

Microsoft says the model can be tried on Arena, Artificial Analysis and MAI Playground. Its Foundry model catalogue for developers, when we read it on 28 October, listed MAI-Image-2.6 and MAI-Image-2.6-Flash, both marked Preview, and no MAI Image 2.7 entry.

The Rest of the Day’s Releases

The same day, Microsoft AI released MAI Cyber 1.1 Flash, which it says delivers “comparable performance to MAI Cyber 1 Flash while costing 18% less per scan on an internal MDASH vulnerability-scanning evaluation suite.” MDASH is Microsoft’s security-scanning harness.

Microsoft says that after a first scan with MAI Cyber 1 Flash, nearly 9 in 10 later scans use the MAI profile rather than the OpenAI-only profile, and that MAI Cyber now accounts for almost 50% of all MDASH scans.

GitHub, which Microsoft owns, said on 27 October that Copilot now supports server-to-server authentication for GitHub Apps installed on organizations, in public preview. Agents built with the Copilot SDK can make requests on behalf of an organization, and usage is billed to the organization that owns the installation.

Connecting the Coverage

In June we covered Microsoft Build, where we reported Microsoft’s homegrown AI models. In July we reported that MAI models were replacing OpenAI and Anthropic inside Excel and Outlook.

This month we reported that Microsoft is building a 2.8T-parameter model, and a day ago that Microsoft’s cyber model doubled to 50% of scans. On image editing, Krea 3 priced edits at $0.06 on 20 October; Krea 3 Ultra sat ninth on the same Arena board.

The Structural Read

As we read it, Arena puts MAI Image 2.7 third, behind OpenAI’s two newest models and ahead of every Google entry on the board.

Inside MDASH, Microsoft says nearly 9 in 10 later scans use the MAI profile rather than the OpenAI-only profile once a customer has tried it.

GitHub’s change bills Copilot SDK agents to the organization that owns the installation, under the organization’s own budgets and spending controls.

Three Implications

FOR MICROSOFT Its image editor ranks third on Arena and its cyber model accounts for almost 50% of MDASH scans, by Microsoft’s count.

FOR OPENAI Its GPT-Image-2.5 Sunburst and Flare are the only two models above MAI Image 2.7 on Arena’s image-edit board, both with preliminary scores.

FOR DEVELOPERS MAI Image 2.7 has no Foundry listing yet; the catalogue lists MAI-Image-2.6 and 2.6-Flash as Preview.

The Business Engineer Lens

This story maps onto the Business Engineer framework The Agentic Harness War.

The framework puts the shift plainly: “The agentic surface — the harness — is becoming where value pools in the entire AI stack.”

Microsoft’s MAI Cyber runs inside its own MDASH harness, and GitHub now lets Copilot SDK agents act for an organization. The framework was published in July 2026 and does not discuss these releases.

What Is Not Established

Arena scores come from crowd votes and move as votes come in. MAI Image 2.7 has 3,402 votes against tens to hundreds of thousands for the models around it, and its margin of plus or minus 10 points overlaps the second- and fourth-placed models.

Microsoft’s post gives no price and no developer access date for MAI Image 2.7. We did not contact Microsoft, OpenAI or Arena.

The Bottom Line

Microsoft AI’s MAI Image 2.7 sat third on Arena’s image-edit board on 28 October at 1471, behind two OpenAI models and 44 points above Google’s best listed model. On 27 October, Microsoft said its cyber model now accounts for almost 50% of MDASH scans.