Ideogram Releases First Open-Weight Image Model With 9.3B Parameters and 2K Native Resolution
Ideogram has released Ideogram 4, a 9.3B parameter open-weight text-to-image model trained from scratch. The model features structured JSON prompting, native 2K resolution output, and ranks as the top open-weight model on Design Arena. Available in fp8 and nf4 quantizations under a non-commercial license.
Ideogram Releases First Open-Weight Image Model With 9.3B Parameters and 2K Native Resolution
Ideogram has released Ideogram 4, a 9.3B parameter open-weight text-to-image model trained entirely from scratch. The model is available in two quantizations: nf4 (CUDA-only, Diffusers-compatible) and fp8 (cross-platform), both under the Ideogram 4 Non-Commercial license.
Architecture and Technical Specifications
Ideogram 4 uses a fully single-stream Diffusion Transformer (DiT) architecture with 34 layers. Unlike traditional text-to-image models, it concatenates text and image tokens into a unified sequence processed through the same transformer, enabling cross-modal interaction at every layer.
The model uses Qwen3-VL-8B-Instruct as its text encoder instead of CLIP or T5. Hidden states are extracted from 13 intermediate layers and concatenated, providing multi-scale semantic features. The model supports resolutions from 256px to 2048px (in multiples of 16) with aspect ratios up to 6:1.
Benchmark Performance
According to Ideogram, the model ranks first among open-weight models on Design Arena, a third-party image generation leaderboard focused on design tasks. On the overall Design Arena board, Ideogram 4 trails only proprietary models from OpenAI (GPT Image) and Google (Gemini).
In ContraLabs' blind typography evaluation with ten professional designers, Ideogram 4 was selected as best 47.9% of the time, ahead of Gemini 3.1 Flash Image Preview (30.0%), FLUX.2 [max] (15.5%), and Grok Imagine 1.0 (15.0%). The same designers rated it 3.55/5 for real client work usability, higher than competing models.
On standard open-source benchmarks, Ideogram claims the model leads all tested models on layout control (7Bench) and delivers better text rendering than larger open-weight alternatives including Qwen-Image (20B), FLUX.2 [dev] (32B), and HunyuanImage 3.0 (80B MoE).
Key Features
The model introduces structured JSON prompting, allowing explicit control over composition, style, lighting, color palette, typography, and spatial layout through bounding-box coordinates. It supports multilingual text rendering and can generate images at native 2K resolution without upscaling.
Inference code is available on GitHub, with model weights hosted on Hugging Face behind a license gate. The model requires authentication via Hugging Face tokens and optionally integrates with Ideogram's hosted "magic prompt" API for prompt expansion and Hive for safety screening.
What This Means
Ideogram 4 represents a significant release in open-weight text-to-image models, particularly for design-focused applications. The 9.3B parameter count makes it substantially smaller than competing open models like FLUX.2 [dev] (32B) while claiming superior performance on design and typography benchmarks. However, the non-commercial license limits production use cases. The structured JSON prompting interface and native high-resolution support address key limitations of previous open-weight image models, though real-world performance will depend on community validation beyond company-provided benchmarks.
Related Articles
Anthropic Releases Claude Fable 5.1, Claims 52.6% on New Terminal-Bench-Science Benchmark
Anthropic released Claude Fable (and Mythos) 5.1, claiming a 52.6% score on the new Terminal-Bench-Science 0.1 benchmark — up sharply from 24.7% for Fable 5. Independent testing shows the model's five reasoning levels produce dramatically different output token counts and costs for identical prompts, ranging from $0.10 to $3.30 per request.
DeepSeek Releases V4-Flash-Vision-Exp, First Multimodal Model in V4 Family
DeepSeek has released DeepSeek-V4-Flash-Vision-Exp, its first experimental multimodal model in the V4 family, adding visual understanding to the V4-Flash architecture. The 305B-parameter model shows substantial gains on multimodal agent benchmarks while holding steady on text-only tasks.
GLM-5.3-Flash Debuts as Zhipu AI's First Natively Multimodal Model, 320B Parameters with 18B Active
Zhipu AI has released GLM-5.3-Flash, the first natively multimodal model in its GLM-5 series, built on a 320B-parameter mixture-of-experts architecture with only 18B active parameters. The company claims it outperforms GLM-5.2 while approaching Claude Opus 4.8 on coding and agentic benchmarks at a fraction of the cost. Unsloth has published quantized GGUF versions for local inference.
Z.ai Launches GLM-5.3-Flash: 1M-Token Context, Image Support, Claimed 10x Cost Cut Over GLM-5.2
Z.ai has released GLM-5.3-Flash, a 320-billion-parameter Mixture-of-Experts model with 18 billion active parameters, a 1-million-token context window, and image input support. The model launched on LM Studio's Bionic platform hours after its official unveiling, with LM Studio claiming it is 9-10x cheaper to run than GLM-5.2.
Comments
Loading...