Ideogram 4.0 Tops Open-Weight Image Models With 1285 Elo
Ideogram AI released its latest image-generation model, "Ideogram 4.0," as open weights on June 3, scoring an Elo rating of 1285 on the community-voted benchmark Image Arena and taking the No. 1 spot in the open-weight category. Its lead over second place reaches 115 points, far outpacing Tencent's HunyuanImage-3.0 and Black Forest Labs' FLUX.2 [dev]. Average generation time is reported at 68.7 seconds. Source
Ideogram is an independent AI company known for accurate in-image "text rendering," arriving at version 4.0 after Ideogram 3.0 (March 2025) and Ideogram 2.0 (August 2024). The official account @ideogram_ai positioned the new model as "the best open image model in the world," stating that users can download the weights, fine-tune, and run it on their own hardware. Distribution spans all Ideogram app plans, the API, and MCP, plus weight releases via Hugging Face (fp8/nf4 checkpoints) and native ComfyUI support. Source
The biggest reason the announcement draws attention is that an "open-weight" model with publicly released weights has shown frontier-class performance in image generation, a field where closed models such as OpenAI's GPT Image series and Google's Imagen/Gemini Image have dominated the top ranks. Arena-style leaderboards compute Elo ratings from human blind votes, and Ideogram 4.0 placed around 1204 on the Text-to-Image Arena (8th overall, 1st among open models) and around 5th on Design Arena, closing in on the closed-model field. Source
On the feature side, highlights include high-precision text rendering (0.97 on X-Omni English OCR), native 2K resolution, native transparency, precise layout control, and support for structured JSON prompts including HEX color palettes and bounding boxes. Parameter scale is reported at 9.3B, and technically it adopts a 34-layer single-stream DiT architecture with Qwen3-VL-8B-Instruct as its text encoder, said to be trained on structured JSON captions. Source
Favorable reactions poured in on X right after release. Many praised the value unique to open weights: the ability to fine-tune and run locally, the absence of rate limits, and data privacy. Source On the practical side, users pointed to usefulness in text, layout, and design tasks, citing "crispy 2K text," "bounding boxes that work," and "improved Arabic text support." Source Compared with closed models such as Nano Banana Pro, GPT Image, and Gemini Image, some assessed it as "No. 1 in open and closing in on Google/OpenAI" and "frontier quality achieved with open weights." Source Few negative opinions were seen at this point, with excitement spreading that it is a big day for the open-source community. Official documentation positions the model around "prompt fidelity, crystal-clear type, reliable editing, native transparency, style control, and design-quality outputs." Source Note that these figures are a snapshot at the time of posting, and Arena rankings may shift with voting.