Ideogram 4.0: status and timeline · The Techno Optimist

Ideogram 4.0

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.

DeployStage 4 of 5

Released June 2026; top open-weight model on Design Arena, behind GPT and Gemini overall.

Updated 3 Jun 2026Checked 11 Oct0 updates this week

Milestones

No announced next step

First public model (0.1)22 Aug 2023Complete.
Ideogram 2.0Aug 2024Complete.
Ideogram 3.0: better realism and complex text layoutsMar 2025Complete.
Ideogram 4.0 with open weights3 Jun 2026Complete.

Most important updates

Ideogram releases 4.0, a 9.3B-parameter image model with open weights3 Jun 2026

Current obstacles

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

Physics limits

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

How it works

4 parts

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

^R reported by the company

Update log

1 update

Wed 3 Jun

Major: Ideogram releases 4.0, a 9.3B-parameter image model with open weightsPressSoftware

About Ideogram

The team behind Ideogram 4.0

Ideogram

Image generation with accurate text, open-weight Ideogram 4.0

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.

ideogram.ai (opens ideogram.ai)Company profile →

Founded: 20224 yrs
Headquarters: Canada
Status: Private
Valuation: private
Raised: $80M1 round
Last round: Series A · $80MFeb 2024
Works in: AIImage generation
Coverage: 1 program · 1 updatelatest 3 Jun 2026checked 11 Oct
Lead investors: Andreessen Horowitz
Investors: Andreessen Horowitz
People: Mohammad NorouziCo-founderWilliam ChanCo-founderChitwan SahariaCo-founderJonathan HoCo-founder

Spec

Ideogram 4.0

Parameters

9.3 billion^R (reported)

Resolution

256 to 2,048 px per side, up to 6:1^R (reported)

Text encoder

Qwen3-VL-8B-Instruct^R (reported)

English text accuracy (X-Omni OCR)

0.97^R (reported)

Typography first-place win rate

47.9% (vs 30.0% Gemini 3.1 Flash Image)^R (reported)

Weights licence

Non-commercial; code Apache 2.0^R (reported)