I don’t think it can solve super difficult math problems, but it is really good and we hope you enjoy it,he wrote.
OpenAI’s own pitch leans on four improvements: faster generation, improved fidelity for more natural and recognizable images, consistent details across multiple edits, and comment-based edits that let users change only what they specify rather than regenerating an entire image.
Rolling Out Everywhere at Once
ChatGPT Images 2.5 is now available to all users across web, desktop, and mobile applications. According to technology journalist David Nield, usage limits vary by plan, though OpenAI has kept the specific numbers vaguely defined.
Free users may also notice a change while waiting for images to generate. OpenAI added a playable Snake game during generation delays, a small addition clearly designed to keep people in the app rather than tabbing away. That convenience comes with a tradeoff: free tier users get fewer images per day, with more generous limits reportedly available during off-peak hours after 9pm.
Where the Sketch Tool Pulls Ahead
The most notable new feature is the Sketch tool, which converts basic scribbles into fully realized images. Nield tested this directly, turning a rough sketch into a finished children’s book illustration. Gemini has no equivalent dedicated sketch tool, though it can use image prompts for similar inspiration.
Editing precision got a real upgrade too. Nield tested both models on a photorealistic cabin scene, and ChatGPT Images 2.5 allowed comments on specific sections, selective erasing, and background changes. It removed furniture and shifted the time of day cleanly, complete with rearranged shadows. Gemini’s editing tools are simpler but still handled basic changes like removing chairs and adjusting lighting effectively.
Food Still Looks Like AI Food
One area where both models still show their limits is food imagery, long known for triggering an uncanny valley effect. Nield tested both tools on a prompt for fast food display images. Both produced edible looking results, but with that familiar generic AI polish that gives them away. ChatGPT’s output leaned slightly more realistic on the food itself, while Gemini handled backgrounds better.
- ChatGPT Images 2.5: Stronger on food realism, dedicated Sketch tool, comment-based editing
- Gemini Nano Banana 2: Stronger backgrounds, simpler but effective editing tools
Close Race, Small Edges
Overall, the two models perform very closely across most tasks. Gemini Nano Banana 2 remains excellent for a wide range of use cases, but ChatGPT Images 2.5 holds a slight edge thanks to its sketching capability and prompt saving templates.
The bigger question isn’t which tool wins on a feature checklist. It’s what this means for the value of human-made art and photography as editing tasks that once took professionals hours now take anyone seconds, with both tools only getting better from here.
Hashlytics Take
The Snake game detail says more about OpenAI’s priorities than the actual image upgrades do. Adding a distraction during generation delays while simultaneously tightening free tier limits is a retention play dressed up as a nice touch. It works because people will wait longer for something entertaining than something boring, which means OpenAI gets to slow-walk free users without the complaints that usually come with a straightforward quota cut. Altman’s math joke is honest framing. The real story here isn’t a leap in image quality, it’s OpenAI tuning the free experience to push more people toward paid plans.
Follow Hashlytics on Bluesky, Facebook, LinkedIn , Telegram and X to Get Instant Updates
