Image Models · checked Oct 17, 2026
Every pixel, ranked.
Live rankings from Artificial Analysis and LMArena, real per-image prices, and what you can run locally. Every number links to its source.
Leaderboards
Each board keeps its own scale. AA rows were parsed from page data and may differ slightly from the rendered page.
Text-to-Image: Artificial Analysis
Swipe the chart sideways for all 20 models →
Nano Banana 2.1
Grok Imagine Image 2.0Nano Banana 2Gemini 3.1 Flash Image
Muse ImageNano Banana ProGemini 3 Pro ImageNano Banana 2 LiteGemini 3.1 Flash Lite Image
Seedream 5.0 Pro
Qwen-Qwen-Qwen-Grok Imagine ImageQualityElo (AA-Image-T2I v2.1) · 20 models · top 5 coloured · axis does not start at zeroSource: artificialanalysis.ai ↗ · checked 2026-10-23
Text-to-Image: LMArena
Swipe the chart sideways for all 20 models →
Grok Imagine Image 2.0Gemini Nano Banana 2.1Grok Imagine Image 2.0build 2026-08-01Muse ImageGemini 3.1 Flash ImageSeedream 5.0 ProQwen-Gemini 3.1 Flash-Gemini 3 Pro Image2KGemini 3 Pro Image PreviewQwen-Arena score · 20 models · top 5 coloured · axis does not start at zeroSource: lmarena.ai ↗ · board dated 2026-10-07, checked 2026-10-23
Image Edit: LMArena
Swipe the chart sideways for all 20 models →
Grok Imagine Image 2.0Gemini Nano Banana 2.1Grok Imagine Image 2.0build 2026-08-01Muse ImageSeedream 5.0 ProGrok Imagine ImageQualityGemini 3 Pro Image2KGemini 3.1 Flash ImageGemini 3 Pro Image PreviewQwen-Arena score · 20 models · top 5 coloured · axis does not start at zeroSource: lmarena.ai ↗ · board dated 2026-10-19, checked 2026-10-23
Score vs price
13 models
Subscription only (Basic $10 to Mega $120 a month). No API, no per-image price, and not ranked on the arenas we track. docs.midjourney.com ↗ checked 2026-10-17
Open models and ComfyUI
| Model | Licence | ComfyUI | Minimum VRAM | Rank |
|---|---|---|---|---|
| FLUX.2 [klein] 4B | Apache 2.0 | native template (klein) | Official: ~13 GB (model card); ComfyUI docs: distilled 8.4 GB, base 9.2 GB on RTX 5090 huggingface.co ↗ docs.comfy.org ↗ | AA #95 (838); LMArena T2I #70 |
| FLUX.2 [dev] 32B | flux-non-commercial-license | native (docs.comfy.org/tutorials/flux/flux-2-dev) | Official: not captured; card points to quantized consumer setups Community report: see modelvram.com calculator huggingface.co ↗ | AA #44 (1000); LMArena T2I #38 (1146) |
| Qwen-Image (2512) / Qwen-Image-Edit 2511 | Apache 2.0 | native; bf16 40.9 GB, fp8 20.4 GB files; 8-step lightx2v LoRA | Official: ComfyUI docs: fp8 uses 86% of a 24 GB RTX 4090D Community report: GGUF quants exist (not verified here) docs.comfy.org ↗ huggingface.co ↗ | LMArena T2I #44 (1125); Edit #37/#39 |
| Z-Image-Turbo (Tongyi) | Apache 2.0 | native template | Official: fits 16 GB consumer GPUs (model card + ComfyUI docs) huggingface.co ↗ docs.comfy.org ↗ | AA #70 (934); LMArena T2I #59 (1083) |
| HiDream-O1-Image | MIT | ComfyUI docs page exists (tutorials/image/hidream/hidream-o1); VRAM not stated | Official: not stated huggingface.co ↗ docs.comfy.org ↗ | LMArena T2I #47 (1116) |
| HunyuanImage 3.0 | tencent-hunyuan-community | not verified | Official: not captured lmarena.ai ↗ | LMArena T2I #35 (1150); Edit (Instruct) #26 |
Sources and gaps
- AA-Image-T2I v2.1 ↗ parsed from page data; 104 models
- LMArena Text-to-Image ↗ · board 2026-10-07 · 6,604,125 votes
- LMArena Image Edit ↗ · board 2026-10-19 · 30,985,932 votes
- AA image editing page not parsed; gap
- Release dates unverified for Ideogram 4.5, Recraft V4.1, Qwen-Image-3.0, Reve 2.1, HiDream-O1
- Midjourney has no API, no per-image price and no arena rank
- No AA image-edit data
- OpenAI and Microsoft per-image costs are estimates from token rates
- Ad use cases are vendor-stated only; no brand case studies verified
- HiDream X handle not found
- AA rows were parsed from embedded page data and may differ slightly from the rendered page (the top 5 match the search snapshot)
FAQ
Which AI image model is best right now?
GPT Image 2.5 Sunburst (max) leads Artificial Analysis text-to-image (1222); LMArena text-to-image (board 2026-10-07) is led by GPT-Image-2.5 Sunburst. Checked 2026-10-17.
What is the cheapest image model with an official price?
Recraft V4.1 (Utility / Pro / Flash) lists $0.007 per image (flash raster) as of 2026-10-17. See its page for batch and other tiers.
Can I run an image model locally?
Yes. 6 open models are listed with licence, ComfyUI support and minimum VRAM (official and community figures labelled) in the open models table.