All tasks / Comparisons
GPT Image 2.5 vs MAI-Image-2.6
Compared only on public benchmark results, across the 2 tasks where both are listed: GPT Image 2.5 is a tier higher in 2 of 2 shared tasks.
GPT Image 2.5
OpenAI's image generation and editing model, offered as Flare (fast default) and Sunburst (precise editing) via API and ChatGPT.
- Vendor
- OpenAI
- Weights
- Closed: API and apps only
- Released
- 8 Sep 2026
- API price
- $0.21 per image (API price listed by Artificial Analysis for "GPT Image 2.5 Sunburst (max)": $210.72 per 1k images; quality "max" at 1024x1024 per AA; OpenAI bills $30 per 1M image output tokens (same rate for Sunburst and Flare), lower quality settings cost less)
- License
- proprietary
- Last verified
- 18 Oct 2026
MAI-Image-2.6
Microsoft AI's in-house image generation and editing model, with a cheaper Flash variant; public preview on Microsoft Foundry.
- Vendor
- Microsoft AI
- Weights
- Closed: API and apps only
- Released
- 10 Aug 2026
- API price
- $0.04 per image (API price listed by Artificial Analysis for "MAI-Image-2.6": $38.9 per 1k images; Microsoft Foundry public preview)
- License
- proprietary
- Last verified
- 18 Oct 2026
Task by task
| Task | GPT Image 2.5 | MAI-Image-2.6 | Higher tier |
|---|---|---|---|
| Image generation Models | S1,220 on Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1) · Cloud | A1,166 on Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1) · Cloud | GPT Image 2.5 |
| Image editing Models | S1,199 on Artificial Analysis Image Editing Arena (AA-Image-Editing v2.1) · Cloud | A1,152 on Artificial Analysis Image Editing Arena (AA-Image-Editing v2.1) · Cloud | GPT Image 2.5 |
A tier spans a range of scores, and small gaps on a leaderboard are often within its margin of error, so only a different tier counts as ahead. How we rank.
Same benchmark, both results
| Benchmark | GPT Image 2.5 | MAI-Image-2.6 | Snapshot |
|---|---|---|---|
| Artificial Analysis Image Editing Arena (AA-Image-Editing v2.1) Elo from pairwise human preference (FLUX.2 [dev] anchored at 1000), higher is better | 1,199 | 1,152 | 18 Oct 2026 |
| Artificial Analysis Text to Image Arena (AA-Image-T2I v2.1) Elo from pairwise human preference (FLUX.2 [dev] anchored at 1000), higher is better | 1,220 | 1,166 | 18 Oct 2026 |
| LMArena Image Edit Arena Arena score (Bradley-Terry rating from pairwise human votes), higher is better | 1,524 | 1,428 | 6 Oct 2026 |
| LMArena Text-to-Image Arena Arena score (Bradley-Terry rating from pairwise human votes), higher is better | 1,425 | 1,332 | 7 Oct 2026 |