GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
GPT-56T 861 —
MUSE-SPK 835 -0.7%
GPT-56SC 828 -5.2%
QWEN-38X 824 —
CL-OP55X 822 —
GROK-46H 822 -5%
GPT-6A 820 —
GLM-5 784 -8.4%
CL-FAB5H 743 -5.6%
KIMI-K3X 742 -8.4%
CL-OP5H 720 -5.8%
CL-OP5X 709 -18%
CL-OP46H 698 -5.9%
CL-OP47H 690 -5.9%
GEM-38FH 677 +0.1%
GEM-37FH 657 -24%
GPT-56S 622 —
CL-OP47 582 -0.7%
GPT-55H 582 —
INKL 531 —
GEM-31P 513 —
GEM-3P 499 —
CL-OP46 496 -0.2%
CL-OP48 490 —
← Back to feed

Microsoft MAI-Image-2.5 Enters Arena Text-to-Image at #3 With Elo 1,254

Microsoft AI shipped MAI-Image-2.5 (Preview) on May 26, 2026, posting 1,254 Elo on the Arena text-to-image leaderboard on day one. That is a 72-point gain over MAI-Image-2, which had been the best Microsoft had put on the board.

The top five on Arena Text-to-Image had been locked to OpenAI and Google DeepMind models since the leaderboard launched. MAI-Image-2.5 breaks that hold, making Microsoft the third distinct lab represented in the upper tier.

Where It Sits

The full current top of the Arena Text-to-Image leaderboard, as of May 26:

RankModelElo
1OpenAI (DALL-E 4 family)~1,350+
2Google DeepMind (Imagen 4 family)~1,300+
3Microsoft AI MAI-Image-2.5 (Preview)1,254

The prior MAI-Image-2, released March 19, 2026, had entered at approximately 1,182 Elo. MAI-Image-2-Efficient — the cost-optimised variant — launched alongside it. Both had placed Microsoft in the #3 model family position on the overall leaderboard. MAI-Image-2.5 now holds that individually.

What Changed

Microsoft has not published a technical deep-dive for 2.5. Based on its evaluation criteria, the model targets photorealistic imagery with accurate skin tones and lighting, reliable text rendering for posters and signage, and expressive range across cinematic and stylised outputs. These were the three pillars of MAI-Image-2; the point gap suggests measurable progress on at least photorealism and text accuracy, where Arena voters have historically concentrated their preferences.

The model is in preview. Arena.ai access is live. MAI Playground and Microsoft Foundry availability is scheduled for next week.

Build Context

The announcement came with a pointed hint: “With Build just a week away, there’s much more to come from the Microsoft AI team.” Microsoft Build 2026 opens June 2. MAI-Image-2.5 appears to be a pre-Build warm-up release, not the main event. Given that MAI-Image-2 had already placed Microsoft in the top three at launch in March and this release extends that lead by 72 points in under ten weeks, the cadence suggests aggressive iteration.

For developers, the Foundry rollout next week matters more than the Arena score. MAI-Image-2 had selective enterprise API access at launch via private preview; 2.5 appears to follow the same staged path.

Competitive Picture

Text-to-image has been a two-lab story at the frontier for the past year: DALL-E variants and Imagen variants trading the top spots. MAI-Image-2.5 at 1,254 is the closest any third lab has come. Recraft v4.1-pro entered Arena on May 22 at a lower tier. HiDream O1-Image, the top open-weight entry, entered the leaderboard last week and sits outside the top five.

Microsoft’s path to #1 on Arena Text-to-Image would require closing more than 100 Elo points against OpenAI. That is the number to watch at Build.