Image AI Models
Every image generation model ranked by blind human-preference votes and what it costs. The value frontier shows which ones are actually worth it.
Arena Elo + pricing from Replicate, fal.ai & makers · reviewed 2026-06-01
The verdict
GPT-Image-2.5 Sunburst is the highest-rated text-to-image model we track on Arena (Nov 4, 2026).
Best value (top quality at each price point): wan2.7-image, seedream-5.0-lite, gemini-2.5-flash-image-preview, seedream-4.5. Best open-weights: Qwen-Image-2.1.
Ranked by the crowd
Quality vs. cost, measured
Ranked by Arena Elo (blind, head-to-head human preference votes), paired with the cheapest published price we can find per model.
| # | Model | Maker | Arena Elo | Cost / image |
|---|---|---|---|---|
| 1 | GPT-Image-2.5 SunburstClosed | OpenAI | 1427±5 | —No price captured |
| 2 | GPT-Image-2.5 FlareClosed | OpenAI | 1401±5 | —No price captured |
| 3 | gpt-image-2 (medium)Value pickClosed | OpenAI | 1381±4 | $0.047Replicate |
| 4 | MAI-Image-2.6Closed | Microsoft AI | 1329±5 | —No price captured |
| 5 | Gemini Nano Banana 2.1Closed | 1325±9 | —No price captured | |
| 6 | Grok Imagine Image 2.0Closed | xAI | 1317±7 | —No price captured |
| 7 | Reve 2.1Closed | Reve | 1300±8 | —No price captured |
| 8 | Grok Imagine Image 2.0 (20260801)Closed | xAI | 1288±5 | —No price captured |
| 9 | Muse ImageClosed | Meta | 1269±5 | —No price captured |
| 10 | reve-2.0Closed | Reve | 1268±6 | —No price captured |
| 11 | Gemini 3.1 Flash Image (nano-banana-2, Web Search)Closed | 1257±4 | —No price captured | |
| 12 | Qwen-Image-3.0-ProClosed | Alibaba | 1254±6 | —No price captured |
| 13 | mai-image-2.5Closed | Microsoft AI | 1253±4 | —No price captured |
| 14 | Seedream 5.0 ProClosed | Bytedance | 1250±4 | —No price captured |
| 15 | Gemini 3.1 Flash-Lite Image (nano-banana-2-lite)Closed | 1249±6 | —No price captured | |
| 16 | gemini-3-pro-image-preview-2k (nano-banana-pro)Closed | 1246±3 | —No price captured | |
| 17 | gpt-image-1.5-high-fidelityClosed | OpenAI | 1236±3 | —$0.013–$0.136 reference · the benchmark doesn't state its quality setting |
| 18 | gemini-3-pro-image-preview (nano-banana-pro)Closed | 1232±5 | —No price captured | |
| 19 | Qwen-Image-2.1Open | Alibaba | 1225±6 | —No price captured |
| 20 | ideogram-4.0-qualityOpen | Ideogram | 1200±4 | —No price captured |
| 21 | Qwen-Image-2.0-Pro (2026-06-22)Closed | Alibaba | 1191±6 | —No price captured |
| 22 | uni-1.1-maxClosed | Luma AI | 1189±6 | —No price captured |
| 23 | mai-image-2Closed | Microsoft AI | 1183±5 | —No price captured |
| 24 | uni-1.1Closed | Luma AI | 1183±4 | —No price captured |
| 25 | grok-imagine-imageClosed | xAI | 1169±3 | —No price captured |
| 26 | recraft-v4.1-utility-proClosed | Recraft | 1168±11 | —No price captured |
| 27 | Cosmos3-Super-Text2Image (Agentic)Open | Nvidia | 1164±7 | —No price captured |
| 28 | flux-2-maxClosed | Black Forest Labs | 1161±4 | —Priced only for other settings |
| 29 | grok-imagine-image-proClosed | xAI | 1161±4 | —No price captured |
| 30 | flux-2-flexClosed | Black Forest Labs | 1156±3 | —No price captured |
| 31 | reve-v1.5Closed | Reve | 1154±4 | —No price captured |
| 32 | flux-2-proClosed | Black Forest Labs | 1153±3 | —Priced only for other settings |
| 33 | hunyuan-image-3.0Open | Tencent | 1151±3 | $0.08Replicate |
| 34 | Cosmos3-Super-Text2ImageOpen | Nvidia | 1151±6 | —No price captured |
| 35 | seedream-4.5Value pickClosed | Bytedance | 1149±3 | $0.04fal.ai |
| 36 | gemini-2.5-flash-image-preview (nano-banana)Value pickClosed | 1149±2 | $0.04fal.aifrom $0.039 | |
| 37 | imagen-ultra-4.0-generate-001Closed | 1146±4 | $0.06Replicate | |
| 38 | flux-2-devOpen | Black Forest Labs | 1145±4 | —No price captured |
| 39 | seedream-5.0-liteValue pickClosed | Bytedance | 1140±3 | $0.035fal.ai |
| 40 | seedream-4-2kClosed | Bytedance | 1139±7 | —No price captured |
| 41 | wan2.6-t2iClosed | Alibaba | 1138±3 | —No price captured |
| 42 | recraft-v4.1-proClosed | Recraft | 1131±10 | —No price captured |
| 43 | imagen-4.0-generate-001Closed | 1128±3 | $0.04Replicate | |
| 44 | krea-2-mediumClosed | Krea | 1128±4 | —No price captured |
| 45 | qwen-image-2512Open | Alibaba | 1125±3 | —No price captured |
| 46 | wan2.5-t2i-previewClosed | Alibaba | 1117±3 | —No price captured |
| 47 | seedream-4-falClosed | Bytedance | 1116±7 | —No price captured |
| 48 | krea-2-largeClosed | Krea | 1115±4 | —No price captured |
| 49 | gpt-image-1Closed | OpenAI | 1115±3 | —No price captured |
| 50 | recraft-v4Closed | Recraft | 1115±3 | $0.04fal.ai |
| 51 | hidream-o1-imageOpen | HiDream | 1115±4 | —No price captured |
| 52 | Krea 2 TurboOpen | Krea | 1114±4 | —No price captured |
| 53 | seedream-4-high-res-falClosed | Bytedance | 1113±3 | —No price captured |
| 54 | gpt-image-1-miniClosed | OpenAI | 1109±3 | —No price captured |
| 55 | wan2.7-image-proClosed | Alibaba | 1103±5 | $0.075fal.aifrom $0.03 |
| 56 | wan2.7-imageValue pickClosed | Alibaba | 1101±5 | $0.03Replicate |
| 57 | Recraft V4.1 FlashClosed | Recraft | 1098±10 | —No price captured |
| 58 | mai-image-1Closed | Microsoft AI | 1092±4 | —No price captured |
| 59 | z-image-turboOpen | Alibaba | 1082±5 | —No price captured |
| 60 | seedream-3Closed | Bytedance | 1082±5 | $0.03Replicate |
| 61 | flux-1-kontext-maxClosed | Black Forest Labs | 1075±3 | $0.08fal.ai |
| 62 | flux-2-klein-9bOpen | Black Forest Labs | 1070±3 | —No price captured |
| 63 | qwen-image-prompt-extendOpen | Alibaba | 1061±3 | —No price captured |
| 64 | flux-1-kontext-proClosed | Black Forest Labs | 1059±3 | $0.04fal.ai |
| 65 | imagen-3.0-generate-002Closed | 1057±3 | $0.05Replicate | |
| 66 | qwen-imageOpen | Alibaba | 1057±3 | —No price captured |
| 67 | ideogram-v3-qualityClosed | Ideogram | 1049±4 | $0.09fal.ai |
| 68 | photonClosed | Luma AI | 1036±4 | $0.03Replicate |
| 69 | p-imageClosed | Pruna | 1032±3 | —No price captured |
| 70 | flux-2-klein-4bOpen | Black Forest Labs | 1030±3 | —No price captured |
| 71 | runway-gen4Closed | Runway | 1024±4 | —No price captured |
| 72 | recraft-v3Closed | Recraft | 1022±4 | $0.04fal.ai |
| 73 | flux-1.1-proClosed | Black Forest Labs | 1016±3 | —No price captured |
| 74 | ideogram-v2Closed | Ideogram | 1013±4 | $0.08fal.ai |
| 75 | lucid-originClosed | Leonardo AI | 1013±3 | —$0.017–$0.076 reference · the quote doesn't state its options |
| 76 | glm-imageOpen | Z.ai | 1012±9 | —No price captured |
| 77 | gemini-2.0-flash-preview-image-generationClosed | 976±3 | —No price captured | |
| 78 | flux-1-dev-fp8Open | Black Forest Labs | 969±4 | —No price captured |
| 79 | dall-e-3Closed | OpenAI | 967±4 | —No price captured |
| 80 | flux-1-kontext-devOpen | Black Forest Labs | 942±4 | —Priced only for other settings |
| 81 | stable-diffusion-v35-largeOpen | Stability AI | 938±4 | $0.065Replicate |
| 82 | bagelOpen | Bytedance | 897±6 | —No price captured |
Prices are shown as each provider publishes them at default settings (per image or per megapixel, not normalized across resolution), from Replicate, fal.ai, and the model makers directly. The Value pick badge marks models where nothing higher-rated costs less. “From” marks a cheaper source than the one shown; a blank cost means no comparable per-image price was found (often token-priced API-only models).
Questions builders actually ask
Real demand, honest answers
How do I train a character LoRA that stays consistent?
Consistency comes from caption discipline more than dataset size: describe what varies (pose, lighting) and stay silent on what should stay fixed (the character's face). A tight, well-captioned set of 15–30 images usually beats a large noisy one. Train, generate a test sheet, and re-caption the failures.
Can I run Stable Diffusion on a 6GB GPU?
Yes for SDXL-class and smaller models like Z-Image, which is why low-VRAM cards stay popular with local builders. FLUX will technically load with offloading but runs are slow and cramped. Treat 6GB as an SDXL/Z-Image card, not a FLUX one.
ROCm or CUDA for ComfyUI?
CUDA (NVIDIA) is the smoother path; most nodes, fine-tunes, and tutorials assume it. ROCm (AMD) works in ComfyUI and has improved, but expect more setup friction and the occasional unsupported node. If you're buying for image generation specifically, NVIDIA saves you time.
FLUX vs SDXL for anime?
SDXL still has the deeper anime fine-tune and LoRA library, so a style checkpoint plus a LoRA often wins for that specific look. FLUX leads on photoreal and prompt adherence but its anime ecosystem is thinner. Pick by the style supply, not the base model's raw quality.
Is a closed model like GPT-IMAGE-2 worth it over local?
If you don't need local control, training, or privacy, a closed model removes the entire GPU and setup problem and often gives a stronger one-shot result. The case for local is control: LoRAs, ControlNet, custom workflows, and no per-image cost. Most builders end up using both for different jobs.
How the ranking is built
Quality is blind, head-to-head human-preference votes from the model arena, not vendor claims. Each model is paired with its published price at default settings, with the source shown. The value frontier flags the models actually worth buying: the ones where nothing higher-rated costs less.
- Quality from blind head-to-head votes, not vendor claims.
- Published price per model at default settings, source shown.
- Value frontier flags the best quality at each price point.
- Updated as new models and prices land.