Nejlepší AI na úpravu obrázků
Body z LMArena · stav k 12. listopadu 2026
64 modelů
| 1 | GPT-Image-2.5 SunburstOpenAI | 1 519±5 | 131 648 |
|---|---|---|---|
| 2 | GPT-Image-2.5 FlareOpenAI | 1 484±4 | 128 720 |
| 3 | GPT-Image-2.6 Sunburst (Medium)OpenAI | 1 476±11 | 3 493 |
| 4 | MAI-Image-2.7-PreviewMicrosoft-ai | 1 471±6 | 27 340 |
| 5 | GPT-Image-2 (Medium)OpenAI | 1 462±3 | 364 695 |
| 6 | GPT-Image-2.6 Flare (Medium)OpenAI | 1 460±10 | 3 534 |
| 7 | Grok Imagine Image 2.0xAI | 1 437±6 | 33 892 |
| 8 | MAI-Image-2.6Microsoft-ai | 1 428±5 | 82 766 |
| 9 | Gemini Nano Banana 2.1Google | 1 427±6 | 12 057 |
| 10 | Grok Imagine Image 2.0 (20260801)xAI | 1 418±4 | 134 617 |
| 11 | Krea 3 UltraKrea | 1 412±12 | 2 616 |
| 12 | Muse ImageMeta | 1 401±4 | 204 077 |
| 13 | MAI-Image-2.5Microsoft-ai | 1 400±4 | 183 296 |
| 14 | Seedream 5.0 ProByteDance | 1 392±3 | 349 293 |
| 15 | Grok Imagine Image (Quality, 20260519)xAI | 1 391±6 | 37 087 |
| 16 | Gemini 3 Pro Image (nano-banana-pro, 2K)Google | 1 391±3 | 682 239 |
| 17 | ChatGPT Image Latest (High Fidelity, 20251216)OpenAI | 1 388±3 | 657 370 |
| 18 | Gemini 3.1 Flash Image (nano-banana-2, Web Search)Google | 1 387±3 | 268 281 |
| 19 | Gemini 3 Pro Image Preview (nano-banana-pro)Google | 1 385±3 | 532 464 |
| 20 | Reve 2.1Reve | 1 373±6 | 19 875 |
| 21 | GPT-Image-1.5 (High Fidelity)OpenAI | 1 367±3 | 682 205 |
| 22 | Qwen-Image-2.1Alibabaopen-source | 1 363±6 | 37 959 |
| 23 | Reve 2.0Reve | 1 359±7 | 18 604 |
| 24 | MiniMax H3-ImageMiniMaxopen-source | 1 359±9 | 5 129 |
| 25 | Ideogram 4.5Ideogram | 1 347±4 | 110 249 |
| 26 | UNI-1.1 MaxLuma-ai | 1 333±5 | 42 191 |
| 27 | Grok Imagine ImagexAI | 1 328±2 | 847 359 |
| 28 | UNI-1.1Luma-ai | 1 316±3 | 248 038 |
| 29 | Gemini 3.1 Flash-Lite Image (nano-banana-2-lite)Google | 1 313±5 | 63 104 |
| 30 | Qwen-Image-2.0-Pro (2026-06-22)Alibaba | 1 303±5 | 54 559 |
| 31 | Seedream 4.5ByteDance | 1 303±2 | 1 357 348 |
| 32 | Wan 2.7 Image ProAlibaba | 1 302±4 | 42 986 |
| 33 | HunyuanImage 3.0 InstructTencentopen-source | 1 301±3 | 322 364 |
| 34 | Wan 2.7 ImageAlibaba | 1 301±4 | 43 793 |
| 35 | Seedream 5.0 LiteByteDance | 1 293±3 | 526 525 |
| 36 | Gemini 2.5 Flash Image Preview (nano-banana)Google | 1 292±2 | 11 419 327 |
| 37 | Seedream 4.0 (2K)ByteDance | 1 267±7 | 208 498 |
| 38 | FLUX.2 [max]Black Forest Labs | 1 262±3 | 367 379 |
| 39 | Reve v1.1Reve | 1 261±3 | 657 727 |
| 40 | Kling Image O1Kling | 1 250±4 | 140 087 |
| 41 | FLUX.2 [pro]Black Forest Labs | 1 247±3 | 669 741 |
| 42 | Qwen-Image-EditAlibabaopen-source | 1 241±3 | 1 999 501 |
| 43 | Reve v1Reve | 1 235±5 | 390 853 |
| 44 | Qwen-Image-Edit (2511)Alibabaopen-source | 1 233±3 | 471 839 |
| 45 | Wan 2.6 ImageAlibaba | 1 231±3 | 708 549 |
| 46 | FLUX.2 [klein] 9BBlack Forest Labsopen-source | 1 224±3 | 528 289 |
| 47 | FLUX.2 [flex]Black Forest Labs | 1 224±3 | 433 538 |
| 48 | FLUX.2 [dev]Black Forest Labsopen-source | 1 224±4 | 247 356 |
| 49 | Seedream 4.0 (High Res, fal)ByteDance | 1 216±3 | 1 211 391 |
| 50 | P-Image-Edit | 1 210±3 | 351 492 |
| 51 | Seedream 4.0 (fal)ByteDance | 1 209±6 | 151 408 |
| 52 | Reve v1.1 FastReve | 1 207±3 | 541 114 |
| 53 | Reve Edit FastReve | 1 198±4 | 227 290 |
| 54 | FLUX.2 [klein] 4BBlack Forest Labsopen-source | 1 187±3 | 529 110 |
| 55 | FLUX.1 Kontext [max]Black Forest Labs | 1 182±3 | 390 748 |
| 56 | Wan 2.5 PreviewAlibaba | 1 180±3 | 699 808 |
| 57 | FLUX.1 Kontext [pro]Black Forest Labs | 1 177±3 | 6 389 439 |
| 58 | FLUX.1 Kontext [dev]Black Forest Labsopen-source | 1 150±4 | 3 630 431 |
| 59 | SeedEdit 3.0ByteDance | 1 139±4 | 4 911 694 |
| 60 | GPT-Image-1OpenAI | 1 139±3 | 2 857 643 |
| 61 | GPT-Image-1 miniOpenAI | 1 124±3 | 671 833 |
| 62 | Gemini 2.0 Flash Preview Image GenerationGoogle | 1 082±2 | 4 933 739 |
| 63 | BAGELByteDanceopen-source | 1 027±6 | 13 481 |
| 64 | Step1X-EditStepFunopen-source | 996±5 | 155 354 |
Úprava dodaného obrázku podle zadání. Člověk dostane dvě odpovědi bez jmen modelů a vybere lepší; z těchto soubojů se počítají body. Rozdíl 100 bodů znamená, že lepší model vyhrává zhruba dva souboje ze tří. Tabulku seřadíte kliknutím na záhlaví sloupce; cena je v dolarech za milion tokenů, vstup / výstup. Vědecké otázky a cenu za úlohu měří OpenRouter na testu GPQA Diamond, rychlost a první token jsou medián poskytovatelů za poslední půlhodinu.
Body: LMArena (licence CC BY 4.0). Ceny, velikost kontextu a data zveřejnění: OpenRouter. Index inteligence: Artificial Analysis přes OpenRouter. Testy s cenou za úlohu a rychlost poskytovatelů: OpenRouter. Datisto modely samo netestuje.