---

LM Arena Text-to-Image Leaderboard

Source: lmarena.ai/leaderboard/text-to-image · 3,918,094 votes · 46 models · Updated Feb 2026

Rank	Model	Organization	Elo Score	CI	License
1	GPT-Image-1.5 High Fidelity	OpenAI	1248	±5	Proprietary
2	Gemini-3-Pro Image Preview 2K	Google	1237	±5	Proprietary
3	Gemini-3-Pro Image Preview	Google	1233	±5	Proprietary
4	Grok Imagine Image	xAI	1174	±6	Proprietary
5	FLUX.2 Max	Black Forest Labs	1169	±4	Proprietary
6	Grok Imagine Image Pro	xAI	1166	±6	Proprietary
7	FLUX.2 Flex	Black Forest Labs	1158	±4	Proprietary
8	Gemini 2.5 Flash Image	Google	1157	±3	Proprietary
9	FLUX.2 Pro	Black Forest Labs	1156	±4	Proprietary
10	HunyuanImage 3.0	Tencent	1151	±3	Community
11	FLUX.2 Dev	Black Forest Labs	1150	±5	Proprietary
12	Imagen Ultra 4.0	Google	1149	±4	Proprietary
13	Seedream 4 2K	ByteDance	1141	±6	Proprietary
14	Seedream 4.5	ByteDance	1141	±4	Proprietary
15	Qwen-Image 2512	Alibaba	1139	±5	Apache 2.0
16	Imagen 4.0	Google	1135	±3	Proprietary
17	Wan2.6 T2I	Alibaba	1126	±6	Proprietary
18	Seedream 4 (Fal)	ByteDance	1119	±6	Proprietary
19	Wan2.5 T2I Preview	Alibaba	1117	±4	Proprietary
20	GPT-Image-1	OpenAI	1115	±3	Proprietary
21	Seedream 4 High-Res	ByteDance	1114	±4	Proprietary
22	GPT-Image-1 Mini	OpenAI	1100	±4	Proprietary
23	MAI-Image-1	Microsoft AI	1094	±4	Proprietary
24	Seedream 3	ByteDance	1084	±5	Proprietary
25	Z-Image Turbo	Alibaba	1083	±7	Apache 2.0
26	FLUX.1 Kontext Max	Black Forest Labs	1076	±3	Proprietary
27	FLUX.2 Klein 9B	Black Forest Labs	1065	±4	Non-Commercial
28	Qwen-Image (Prompt Extend)	Alibaba	1061	±3	Apache 2.0
29	FLUX.1 Kontext Pro	Black Forest Labs	1060	±3	Proprietary
30	Imagen 3.0 (002)	Google	1059	±3	Proprietary
31	Qwen-Image	Alibaba	1058	±2	Apache 2.0
32	P-Image	Pruna	1053	±5	Proprietary
33	Ideogram v3 Quality	Ideogram	1050	±4	Proprietary
34	Photon	Luma AI	1037	±4	Proprietary
35	FLUX.2 Klein 4B	Black Forest Labs	1021	±4	Apache 2.0
36	Recraft v3	Recraft	1021	±3	Proprietary
37	FLUX 1.1 Pro	Black Forest Labs	1017	±3	Proprietary
38	Lucid Origin	Leonardo AI	1015	±3	Proprietary
39	Ideogram v2	Ideogram	1015	±3	Proprietary
40	GLM-Image	Z.ai	1013	±9	MIT
41	Gemini 2.0 Flash Image	Google	976	±3	Proprietary
42	FLUX.1 Dev FP8	Black Forest Labs	970	±4	Open
43	DALL·E 3	OpenAI	969	±4	Proprietary
44	FLUX.1 Kontext Dev	Black Forest Labs	942	±4	Non-Commercial
45	Stable Diffusion v3.5 Large	Stability AI	939	±4	Open
46	BAGEL	ByteDance	900	±6	Apache 2.0

---

Artificial Analysis Text-to-Image Arena

Source: artificialanalysis.ai/text-to-image/arena · 15 specialized visual categories

Category	Description
General & Photorealistic	Overall image quality and photorealism
Anime	Japanese animation-style generation
Text & Typography	Accuracy of rendered text in images
People: Portraits	Single-subject human portrait quality
People: Groups & Activities	Multi-person scenes and interactions
Nature & Landscapes	Environmental and natural scenes
Traditional Art	Painting, watercolor, and fine art styles
Cartoon & Illustration	Stylized cartoon and illustration
Vintage & Retro	Period-accurate retro aesthetic
Futuristic & Sci-Fi	Science fiction and futuristic imagery
Graphic Design & Digital Rendering	Design-oriented visual outputs
Fantasy & Mythical	Fantasy creatures and mythology
UI/UX Design	User interface mockups
Commercial	Product photography and marketing imagery
Physical Spaces	Architectural and interior rendering

Recently added models (Feb 2026): Grok Imagine Image Pro, Z-Image Base, Qwen Image Plus 2601, HunyuanImage 3.0 Instruct, Wan2.6 T2I, FLUX.2 Klein (9B & 4B), GLM-Image.

---

Key Observations

Insight	Detail
Proprietary Dominance	Top 5 positions held exclusively by OpenAI, Google, and xAI; highest open-weight model (Qwen-Image, Apache 2.0) at rank 15
Rapid Capability Growth	GPT-Image-1.5 (Elo 1248) vs. GPT-Image-1 (Elo 1115) — 133 Elo improvement within a single model family
Chinese Model Competitiveness	Alibaba (Qwen-Image, Wan2.6, Z-Image), ByteDance (Seedream 4.5), and Tencent (HunyuanImage 3.0) in top-20
Open-Weight Progress	Qwen-Image (Apache 2.0, Elo 1139) and FLUX.2 Klein 4B (Apache 2.0, Elo 1021) demonstrate open models closing the gap
FLUX Ecosystem	Black Forest Labs holds 9 of 46 positions spanning Max, Pro, Flex, Dev, Klein, and Kontext variants

---

Quantitative Metrics Reference

Metric	Full Name	Description
FID	Frechet Inception Distance	Distributional similarity between generated and real images
IS	Inception Score	Quality and diversity of generated images
CLIP Score	CLIP-based Alignment Score	Text-image semantic alignment via CLIP embeddings
VQAScore	VQA-based Score	Compositional evaluation via visual question answering
HPSv2	Human Preference Score v2	Human preference prediction for T2I outputs
ImageReward	Image Reward	Learned human preference model
LPIPS	Learned Perceptual Image Patch Similarity	Perceptual similarity using deep features
TIFA	Text-to-Image Faithfulness Assessment	Faithfulness evaluation via QA
DSG	Davidsonian Scene Graph	Fine-grained compositional evaluation
DreamSim	Dream Similarity	Human visual similarity judgment
SSIM	Structural Similarity Index	Structural image similarity

---