# Ideogram 4.0 by Ideogram

Canonical page: https://technooptimist.io/ai/ideogram-4
Frontier: [AI](https://technooptimist.io/ai.md) · Field: Image generation · Company: [Ideogram](https://technooptimist.io/companies/ideogram.md)
Stage: Deploy (4 of 5: Research → Proto → Pilot → Deploy → Scaling)
Updated: last update 2026-06-03, facts checked 2026-10-11
Goal: Make an image model a designer can hand real work to: posters, ads and packaging with correctly spelled text in the right place, and weights others can run and build on.
Status: Released June 2026; top open-weight model on Design Arena, behind GPT and Gemini overall.

## Milestones

| Status | Milestone | Date | Slip |
| --- | --- | --- | --- |
| Done | First public model (0.1) | 22 Aug 2023 |   |
| Done | Ideogram 2.0 | Aug 2024 |   |
| Done | Ideogram 3.0: better realism and complex text layouts | Mar 2025 |   |
| Done | Ideogram 4.0 with open weights | 3 Jun 2026 |   |

## Most important updates

- **3 Jun 2026**: Ideogram releases 4.0, a 9.3B-parameter image model with open weights. Ideogram published the weights and inference code of Ideogram 4.0, a 9.3-billion-parameter text-to-image model trained from scratch. ([our story](https://technooptimist.io/stories/ideogram-releases-4-0-a-9-3b-parameter-image-model-with-open-weights-bfd615fa.md) · [original source](https://startupfortune.com/ideogram-40-turns-open-weights-into-a-startup-weapon/))

## Current obstacles

- **Licence limits adoption**: Weights are non-commercial, so companies must still buy a licence or use the API to ship products with it.
- **Self-reported rankings**: Most benchmark wins come from Ideogram's model card; community testing was still under way at launch.

## Physics limits

- **Letters are tiny, exact details**: A single wrong stroke ruins a word, while images tolerate small errors elsewhere; text demands far more precision per pixel.
- **Step-by-step generation costs time**: Diffusion models refine an image over many passes, so quality trades off against speed and cost per picture.

## How it works

### Denoise: From noise to picture

A diffusion transformer starts from random noise and, step by step, shapes it into an image matching the prompt.

### One stream: Words and pixels together

Text tokens and image patches share one sequence, so every layer can check the image against the exact wording.

### Reading: A vision-language encoder

Qwen3-VL-8B reads the prompt; its outputs from 13 layers are combined to give the image model rich detail.

### Layout: Boxes and palettes

Prompts can be JSON giving boxes for each element and hex colours, which the model learned to respect.

## Spec sheet

| Spec | Ideogram 4.0 | Provenance |
| --- | --- | --- |
| Parameters | 9.3 billion | reported |
| Resolution | 256 to 2,048 px per side, up to 6:1 | reported |
| Text encoder | Qwen3-VL-8B-Instruct | reported |
| English text accuracy (X-Omni OCR) | 0.97 | reported |
| Typography first-place win rate | 47.9% (vs 30.0% Gemini 3.1 Flash Image) | reported |
| Weights licence | Non-commercial; code Apache 2.0 | reported |

Provenance: reported = stated by the company; estimated = our estimate; sample = a sample figure.

## About Ideogram

Ideogram makes image models known for getting text inside pictures right, for posters, logos and packaging. Founded in Toronto by former Google researchers; Ideogram 4.0 (June 2026) is its first open-weight model.

- Built its name on legible lettering in images, long a weak spot of image models.
- Ideogram 4.0: 9.3B parameters, weights public for non-commercial use, runs on one 24 GB GPU.
- Raised $80M Series A in Feb 2024 led by Andreessen Horowitz.

- Founded: 2022
- Headquarters: CA
- Status: private
- Raised: $80M
- Last round: Series A $80M, Feb 2024, led by Andreessen Horowitz
- Website: https://ideogram.ai
- People: Mohammad Norouzi (Co-founder); William Chan (Co-founder); Chitwan Saharia (Co-founder); Jonathan Ho (Co-founder)
- Partners, customers, investors: Andreessen Horowitz (investor)

Company page: https://technooptimist.io/companies/ideogram.md

## Related programs

Everything in AI: https://technooptimist.io/ai.md

---

The Techno Optimist tracks every frontier of technology, program by program: https://technooptimist.io/ (a Markdown version of any page: add .md to its URL; site map for agents: https://technooptimist.io/llms.txt).
Data API (JSON, realtime stream): https://technooptimist.io/developers. People read 5 pages free, then Reader $10/month; these Markdown pages are open to agents.