RXed AI News

AI news with a spine.
SCORE 6.5/1011/20 elements Audited 2026-07-24 · RXed table v1.0

Ideogram

Still the model that renders text correctly — and with 4.0 it added open weights, a native MCP server and a $0.03/image API. Buy it for on-brand posters, logos and packaging copy; skip it for video, vectors or anything that needs run-level observability.

SpecialistGenerative ArtMarketingFreemium
Vendor
Ideogram AI Inc. · ideogram.ai
Origin
Canada — Toronto
Pricing
Free (slow credits, public JPG outputs) · Basic ~€6 ($7) · Plus ~€14-18 ($16-20) · Pro ~€44-55 ($48-60) · Team ~€28/user ($30) per month, annual ~20-30% off · API $0.03-0.10 per image (~€0.03-0.09) · Enterprise custom + open-weight self-host commercial license
Users (official only)
Not disclosed recently; independent trackers reported ~1.2M monthly active and ~150k paying subscribers (~$20M ARR) as of Q3 2024 (source, 2024-09-30)

Element scores

Reactive
Retrieval & Memory
Orchestration
Validation
Models
Primitives
Pr8
Prompts
Em
Embeddings
Cx7.5
Context
Tr4
Tracing
Lg
LLM
Compositions
Fc5
Function calling
Vx
Vector store
Rg
RAG
Gr5.5
Guardrails
Mm7
Multimodal
Deployment
Ag3.5
Agents
Ft8
Fine-tuning
Fw7
Frameworks & harnesses
Ev
Evaluations
Sm7.5
Small models
Emerging
Ma
Multi-agent
Sy
Synthetic data
Pc8
Protocols
In
Interpretability
Th
Thinking models
Tap or hover any element to see why it got that score.

Strengths

Text rendering is the moat: Ideogram 4.0 (03/06/2026) scores 0.97 on X-Omni English OCR and ranks first among open-weight models on the Arena.ai and DesignArena text-to-image leaderboards, ninth overall behind only closed models from OpenAI and Google. It is a 9.3B single-stream Diffusion Transformer with a Qwen3-VL-8B text encoder, native 2048px output, JSON layout control with bounding boxes, and it is Ideogram's first open-weight foundation model — weights on Hugging Face and GitHub, the nf4 checkpoint fitting a single 24 GB GPU. On top of that sits a full editing suite (reframe, background replace, face swap, upscale, describe), custom brand-model training, a cheap predictable API ($0.03-0.10/image), and an official MCP server on every plan that makes it the best-connected image tool for agent pipelines.

Honest dings

It is image-only: no video, no audio and no editable SVG vectors, so a design pipeline still needs other tools for motion and true vector work. Observability is thin — history and credit counts, but no run traces or cost attribution. The 'open weights' come with a catch: the inference code is Apache-2.0 but the weights sit under a Non-Commercial Model Agreement, so any revenue use needs the paid API, subscription or an enterprise license — and there is no IP indemnification on the open path. Free-tier images are public. Recent user and revenue figures are not officially disclosed.

Best for: Designers, marketers and automation builders who need brand-consistent logos, posters, packaging and ad creative with legible in-image text at a predictable per-image cost — especially agent pipelines that want image generation over MCP, or enterprises that need to fine-tune and run the model behind their own firewall.

Sources

Every audit lists the research it rests on — transparency and traceability are the product. Tools evolve: each audit is a snapshot of its audit date, and re-audits supersede older versions (kept below for reference).

  • ideogram.ai/blog/ideogram-4.0 — Official technical release (03/06/2026): 9.3B single-stream DiT, frozen Qwen3-VL-8B text encoder, structured JSON prompts, first open-weight foundation model (weights on Hugging Face, code on GitHub) (accessed 2026-07-24)
  • developer.ideogram.ai/ideogram-api/api-overview — Official API docs: generate/remix/edit/reframe/replace-background/face-swap endpoints, custom models, MCP server at developer.ideogram.ai/_mcp/server, capabilities across app/API/MCP/open weights (accessed 2026-07-24)
  • ideogram.ai/features/mcp — Official MCP server (mcp.ideogram.ai/mcp): OAuth, streamable HTTP, available on all plans; tool catalog (generate_image, generate_images_bulk, reframe_image, upscale_image, create_collection, train) (accessed 2026-07-24)
  • releasebot.io/updates/ideogram — Official 4.0 open-weight statement + API pricing (Turbo $0.03 / Default $0.06 / Quality $0.10) + MCP-on-all-plans announcement (June 2026) (accessed 2026-07-24)
  • theplanettools.ai/blog/ideogram-4-0-open-weight-lay… — Independent hands-on (verified 07/06/2026): 0.69 mIoU on 7Bench, 0.97 English OCR, nf4 runs on a 24 GB GPU, fp8/nf4 checkpoints, weights cover research/fine-tune but commercial deployment needs a paid license (accessed 2026-07-24)
  • aivideobootcamp.com/blog/ideogram-4-0-complete-guid… — Independent: model history V1-V4, first frontier-grade open-weight release, Apache-2.0 inference code vs Non-Commercial Model Agreement weights, no IP indemnification on the open path, subscription+API+fal.ai revenue paths (accessed 2026-07-24)
  • pricepertoken.com/ideogram-pricing — Independent pricing verified June 2026: Plus $20/mo (~1,000 images on 3.0 Default), Pro $60/mo, Team $30/user, annual ~25-30% off, free tier public JPG-only (accessed 2026-07-24)
  • eesel.ai/blog/ideogram-pricing — Independent API per-image pricing table (4.0 Turbo $0.03 -> Quality $0.10; 3.0 tiers; character-reference surcharge; utility endpoints for remove-bg/upscale/describe) (accessed 2026-07-24)
  • tracxn.com/d/companies/ideogram/__FH8PLnWCb6_QLGv41… — Company profile: founded 2022 in Toronto, Canada by ex-Google researchers; $96.5M raised over 2 rounds ($80M Series A led by Andreessen Horowitz, Feb 2024); ~67 employees as of Jun 2026 (accessed 2026-07-24)
  • zipdo.co/ideogram-statistics — Independent stats aggregation: ~1.2M MAU and 150k paid subscribers, ~$20M ARR, 5M registered users as of Q3/Q4 2024 (no more recent official figure available) (accessed 2026-07-24)