AI image editors ranked on changing one detail
See how the image-editing leaderboard changes when a generator must keep the rest of a source photo intact.

👁 111 reads
A furniture store already has a photograph of the chair. The assignment is to turn the upholstery blue while preserving the grain on the walnut legs. I would take this job to the editing table first. A fresh rendering of a beautiful chair could still require another product shoot.
What the board measures
Microsoft’s MAI Image 2.7 leads this snapshot. Its narrow advantage over the next row is a reason to look closely at reference fidelity on your own materials. Pair a high-contrast change with a deliberately untouched region, then inspect both at the source resolution.
The November 17, 2026 snapshot
The table covers the top 25 and additional recently added entries captured in the source. Prices are the evaluator’s reported USD benchmark estimates. A blank price means the evaluator did not supply one.
| Rank and model | Elo and votes | Benchmark price |
|---|---|---|
| 1. MAI-Image-2.7 | 1215 (5,478 votes) | $34.82 /1k imgs |
| 2. GPT Image 2.6 Sunburst (max) | 1211 (6,554 votes) | $210.72 /1k imgs |
| 3. GPT Image 2.6 Flare (max) | 1205 (6,798 votes) | $210.72 /1k imgs |
| 4. GPT Image 2.5 Sunburst (max) | 1202 (18,971 votes) | $210.72 /1k imgs |
| 5. GPT Image 2.5 Flare (max) | 1185 (19,853 votes) | $210.72 /1k imgs |
| 6. Kling Image 4.0 | 1183 (5,628 votes) | $66.00 /1k imgs |
| 7. MAI-Image-2.7-Flash | 1179 (5,891 votes) | $10.24 /1k imgs |
| 8. Seedream 5.5 | 1160 (5,101 votes) | $28.00 /1k imgs |
| 9. Krea 3 (Ultra) | 1154 (6,191 votes) | $100.00 /1k imgs |
| 10. Nano Banana 2.1 | 1152 (7,606 votes) | $33.60 /1k imgs |
| 11. MAI-Image-2.6 | 1150 (17,367 votes) | $38.91 /1k imgs |
| 12. MAI-Image-2.6-Flash | 1145 (11,493 votes) | $19.46 /1k imgs |
| 13. GPT Image 2 (high) | 1140 (15,237 votes) | $211.00 /1k imgs |
| 14. H3-Image | 1135 (3,267 votes) | $26.54 /1k imgs |
| 15. H3-Image Highspeed | 1132 (3,230 votes) | $45.25 /1k imgs |
| 16. Seedream 5.0 Pro | 1132 (14,130 votes) | $90.00 /1k imgs |
| 17. Muse Image | 1129 (14,049 votes) | $10.00 /1k imgs |
| 18. Nano Banana 2 (Gemini 3.1 Flash Image) | 1126 (17,364 votes) | $67.00 /1k imgs |
| 19. MAI-Image-2.5-Pro | 1124 (16,380 votes) | $108.54 /1k imgs |
| 20. MAI-Image-2.5 | 1124 (14,227 votes) | $48.13 /1k imgs |
| 21. Krea 3 (Fast) | 1122 (5,927 votes) | $70.00 /1k imgs |
| 22. FLUX 3 Image | 1119 (5,630 votes) | $48.00 /1k imgs |
| 23. GPT Image 1.5 (high) | 1118 (14,184 votes) | $133.00 /1k imgs |
| 24. Grok Imagine Image 2.0 | 1118 (9,414 votes) | $60.00 /1k imgs |
| 25. Nano Banana Pro (Gemini 3 Pro Image) | 1111 (16,251 votes) | $134.00 /1k imgs |
| 26. HiDream-O1-Edit-2.0 | 1109 (6,049 votes) | $75.00 /1k imgs |
| 48. ADM 3 Cloud (Image) | 1040 (3,578 votes) | Coming soon |
Reading the result
Preference and cost. The score comes from the listening or viewing comparison. The price column describes a separate estimate. For a working budget I would record actual spend across all attempts and divide by the number of accepted deliverables.
For the next step, see the Quantslant earlier MAI image coverage.
Arena peer comparison
Arena uses its own matchups and snapshot. Keep these scores separate from the Artificial Analysis table.
| Model | Arena placement | Reported score |
|---|---|---|
| GPT-Image-2.5 Sunburst | 1 | 1519 |
| GPT-Image-2.5 Flare | 2 | 1484 |
| GPT-Image-2.6 Sunburst (Medium) | 3 | 1481 |
| MAI-Image-2.7 | 4 | 1475 |
| GPT-Image-2 (Medium) | 5 | 1462 |
| GPT-Image-2.6 Flare (Medium) | 6 | 1461 |
| Grok Imagine Image 2.0 | 7 | 1437 |
| MAI-Image-2.6 | 8 | 1427 |
| Gemini Nano Banana 2.1 | 9 | 1427 |
| Grok Imagine Image 2.0 (20260801) | 10 | 1419 |
| Krea 3 Ultra | 11 | 1412 |
| MAI-Image-2.5 | 12 | 1400 |
| Muse Image | 13 | 1400 |
| Seedream 5.0 Pro | 14 | 1392 |
| Grok Imagine Image (Quality, 20260519) | 15 | 1391 |
| Gemini 3 Pro Image (nano-banana-pro, 2K) | 16 | 1391 |
| ChatGPT Image Latest (High Fidelity, 20251216) | 17 | 1387 |
| Gemini 3.1 Flash Image (nano-banana-2, Web Search) | 18 | 1387 |
| Gemini 3 Pro Image Preview (nano-banana-pro) | 19 | 1385 |
| Reve 2.1 | 20 | 1373 |
| GPT-Image-1.5 (High Fidelity) | 21 | 1367 |
| Qwen-Image-2.1 | 22 | 1363 |
| Reve 2.0 | 23 | 1359 |
| MiniMax H3-Image | 24 | 1358 |
| Ideogram 4.5 | 25 | 1347 |
What is the best AI image edit model? MAI Image 2.7 led the Artificial Analysis editing board on November 17 at 1,215 Elo. For product work, compare its result with the actual source to see whether the untouched material survives.
Editing with more than one reference image
A leatherworker has a photograph of a bag and a separate close-up of its clasp. A model can score well at changing one photograph and still replace the handle when it borrows the clasp from a second image. Arena measures this multi-input operation on its own board. The table below preserves the top 25 as displayed on November 16, read November 17. The votes and uncertainty belong to the named configuration.
| Place | Configuration | Arena score | Votes |
|---|---|---|---|
| 1 | GPT-Image-2.5 Sunburst | 1527 ±5 (preliminary) | 70,812 |
| 2 | GPT-Image-2.5 Flare | 1475 ±5 (preliminary) | 69,942 |
| 3 | MAI-Image-2.7 | 1458 ±10 | 3,647 |
| 4 | GPT-Image-2 (Medium) | 1455 ±3 | 167,372 |
| 5 | GPT-Image-2.6 Flare (Medium) | 1442 ±7 (preliminary) | 10,635 |
| 6 | Gemini Nano Banana 2.1 | 1432 ±7 (preliminary) | 9,631 |
| 7 | Krea 3 Ultra | 1431 ±7 (preliminary) | 11,506 |
| 8 | Seedream 5.0 Pro | 1412 ±4 | 119,364 |
| 9 | Muse Image | 1400 ±4 | 129,822 |
| 10 | Gemini 3.1 Flash Image (nano-banana-2, Web Search) | 1372 ±3 | 146,492 |
| 11 | MiniMax H3-Image | 1367 ±8 | 6,296 |
| 12 | Gemini 3 Pro Image Preview (nano-banana-pro) | 1366 ±4 | 293,627 |
| 13 | Gemini 3 Pro Image (nano-banana-pro, 2K) | 1364 ±3 | 278,901 |
| 14 | ChatGPT Image Latest (High Fidelity, 20251216) | 1350 ±3 | 278,465 |
| 15 | Reve 2.0 | 1346 ±7 | 9,927 |
| 16 | GPT-Image-1.5 (High Fidelity) | 1342 ±3 | 245,850 |
| 17 | Ideogram 4.5 | 1335 ±5 (preliminary) | 26,104 |
| 18 | UNI-1.1 Max | 1308 ±5 | 21,241 |
| 19 | Seedream 4.5 | 1290 ±3 | 621,695 |
| 20 | Gemini 3.1 Flash-Lite Image (nano-banana-2-lite) | 1290 ±5 | 28,298 |
| 21 | Wan 2.7 Image Pro | 1290 ±5 | 17,515 |
| 22 | UNI-1.1 | 1287 ±4 | 114,816 |
| 23 | Wan 2.7 Image | 1280 ±5 | 18,777 |
| 24 | Seedream 5.0 Lite | 1266 ±3 | 405,500 |
| 25 | Reve v1.1 | 1262 ±4 | 264,754 |
The top two entries are different GPT Image 2.5 configurations. Lower in the same table, the observed score of 1,290 is shared by Seedream 4.5, Gemini 3.1 Flash-Lite Image and Wan 2.7 Image Pro. I would use the scores and vote counts together when deciding which models merit a trial. The November source check identified four recent appearances among these 25: MAI Image 2.7, GPT Image 2.6 Flare (Medium), Krea 3 Ultra and MiniMax H3-Image. First observed in a snapshot and released by the vendor are separate dates.
01GPT Image 2.5: archived Sunburst and Flare comparisonThe original GPT Image 2.5 comparison board, current documentation and a clear separation from the newer 2.6 comparison. Read the limits of the evidence.Read →
02MAI-Image-2.6 Review: Pricing, Availability & Test ResultsMicrosoft's MAI-Image-2.6 ranks 5th on Artificial Analysis. We tested both versions: quality scores, speed, pricing, and where you can use them.Read →
03MiniMax H3-Image: Price, API, Test Results and ComparisonMiniMax H3-Image in one place: what it does, where to use it, what it costs, the licence and my test of 170+ images against Seedream 5.5, Nano Banana 2.1 and GPT-Image-2.6.Read →
04FLUX 3 Image: Native 4K, Bounding Boxes & Multi-EditBlack Forest Labs released FLUX 3 Image on October 1, 2026: native 4K up to 5456×3072px, bounding box composition, and multi-area editing without touching the rest.Read →
05Ideogram 4.5: repeated-edit tests, pricing and image examplesExplore Ideogram 4.5 editing tests, source images, USD pricing and 57 gallery outputs, with clear distinctions between quality and edit precision.Read →Sources
- https://artificialanalysis.ai/image/leaderboard/editing (accessed November 17, 2026)
- https://arena.ai/leaderboard/image/image-edit/single-image (accessed November 17, 2026)
- https://arena.ai/leaderboard/image/image-edit/multi-image (accessed November 17, 2026)
- https://arena.ai/company/leaderboard-changelog (accessed November 17, 2026)
- https://huggingface.co/datasets/lmarena-ai/leaderboard-dataset/tree/main/image_edit (accessed November 17, 2026)
More on this from QUANTSLANT
benchmarks·ACE Studio models, access and benchmark guideACE Studio model and access guide with a practical US-English evaluation brief and links to publisher materials.