← AI Tools

Explainer

ChatGPT Images 2.5: what’s new, what’s free, and how to get better pictures

OpenAI’s newest image model arrived in ChatGPT on September 8, 2026. It’s faster, keeps people looking like themselves, and changes only what you ask it to. Here’s what’s new, what the free plan gets, and how to use the new tools.

Short answer

ChatGPT Images 2.5 launched September 8, 2026 for every plan. It generates up to 50% faster than Images 2.0, keeps faces from your photos looking like you, and edits only what you ask. Free users get it, with a small daily limit OpenAI doesn’t publish (users report about 2–3 images a day). “Images with thinking” needs Plus or higher.

What’s new in Images 2.5

Free vs paid: who gets what

PlanImage creationImages with thinking
FreeLimited (number not published)No
Go ($8)More than FreeNo
Plus ($20)More complex and accurate imagesYes
Pro (from $100)Unlimited and faster (within abuse guardrails)Yes

From OpenAI’s pricing page and its “Images in ChatGPT” help article, checked Oct 12, 2026. Free users report about 2–3 images per 24 hours reported.

Want more pictures without paying? Other assistants have their own free allowances; see which AI to use for images. Not sure Plus is worth it? Free vs Go vs Plus.

How to use the new features

Prompts that get better pictures

  1. Say what it’s for: “a poster for a school bake sale, to print on letter paper” beats “a poster”.
  2. Describe, then constrain: subject, setting, style, lighting, then what must not change or appear.
  3. Put exact text in quotes: “The headline reads ‘Bake Sale Friday’”. Keep it short; small text is still the weakest point.
  4. Edit one thing at a time. Several changes in one request are more likely to drift.

The same habits work for any AI; more in how to ask AI better questions.

Watermarks and AI labels

Images from ChatGPT carry C2PA Content Credentials (metadata saying an AI made it) and an invisible SynthID watermark. Anyone can check an image at openai.com/verify. Taking a screenshot strips the metadata, but the watermark usually survives light edits and cropping. OpenAI notes these signals don’t prove an image is accurate or who owns it. Under OpenAI’s terms, you own what you create, but the result may not be unique.

Weak spots to know about

Sources: OpenAI: introducing ChatGPT Images 2.5 (Sept 8, 2026) · OpenAI: images in ChatGPT · OpenAI: provenance signals · ChatGPT pricing · OpenAI: terms of use · Tom’s Guide: free-tier image limits

Quick answers

Is ChatGPT Images 2.5 free?

Yes. OpenAI says ChatGPT Images is available on all plans, including Free. The free plan has a small daily limit that OpenAI doesn’t publish; ChatGPT tells you when you’ve reached it. Users report roughly 2–3 images a day on Free.

How many images can I make with ChatGPT for free?

OpenAI doesn’t publish a number; the pricing page only says “limited image creation” for Free and “more” for Go. Users report about 2–3 images per 24 hours on the free plan, and the limit can change with demand.

What is “images with thinking”?

A mode where ChatGPT reasons about the picture before drawing it, which helps with diagrams, layouts and text in images. It’s available on Plus, Pro and Business, not on Free or Go.

Can people tell an image was made with ChatGPT?

Often, yes. OpenAI adds C2PA Content Credentials and an invisible SynthID watermark to images from ChatGPT and its API. Anyone can upload an image to openai.com/verify to check. Screenshots remove the metadata, but the watermark usually survives.

How do I edit just part of an image?

Open the image, choose Select, highlight the area and describe the change. OpenAI warns that edits can spill beyond the area you selected, so check the whole image afterward.

Keep learning