OpenAI Beefs Up ChatGPT's Image Generation Model
OpenAI launched ChatGPT Images 2.0 on Tuesday, bringing faster generation, multi-image outputs, and sharply improved text rendering to users worldwide. The upgrade matters because it directly targets the weaknesses that have kept AI image generation out of professional workflows,...
Reece Rogers, writing for Wired, put the new model through its paces and found real, measurable progress in image quality and text rendering, alongside one stubborn limitation that OpenAI has not yet cracked. Rogers tested ChatGPT Images 2.0 for several hours and came away generally impressed, with enough caveats to keep the verdict honest.
Why This Matters
OpenAI is not running a charity here. The company has explicitly adopted what it internally calls a "code red" competitive posture, accelerating releases across every product category to hold ground against Google, Anthropic, and a crowded field of challengers. Image generation has become a measurable user-retention tool, as last year's viral figurine trend from Google's Nano Banana model demonstrated, and ChatGPT's own caricature wave earlier this year proved the same point. A model that generates images up to 4 times faster than its predecessor, handles multi-image outputs, and produces clean readable text inside images is a direct bid for the professional and semi-professional creative market, not just the hobbyist crowd.
Daily briefing from 50+ sources. Free, 5-minute read.
The Full Story
OpenAI released ChatGPT Images 2.0 on Tuesday, making it available globally to ChatGPT and Codex users, with a more capable version reserved for paying subscribers. The model is also accessible through the API under the name GPT Image 1.5, which means developers can start building it into their own applications immediately.
The single most interesting technical detail is that Images 2.0 connects to ChatGPT's reasoning capabilities, which lets it search the internet for current information before generating an image. That changes what the model can actually do. Rogers tested this by asking the model to generate an infographic showing San Francisco's weather forecast for the following day along with recommended activities. The result included accurate weather conditions for a rainy day and recognizable renderings of the Ferry Building, Castro Theater, Painted Ladies, and the Transamerica Pyramid. That kind of grounded, real-world accuracy was not something earlier image models could pull off reliably.
Images 2.0 also ships with a knowledge cutoff of December 2025, a meaningful update that brings the model's awareness of recent events closer to the present than most competing systems. And because it can run multiple reasoning steps before producing output, the model can generate more than one image from a single prompt, including multi-page outputs like study booklets or document sets.
Customization gets a boost as well. Users can now specify aspect ratios ranging from 3:1 wide-format to 1:3 tall-format, and the sizing parameters can be adjusted directly in the prompt. That flexibility matters for designers and content creators who need outputs that fit specific platform requirements without post-production cropping.
Text rendering inside images has historically been a weak point across every major AI image system. Rogers notes that just two years ago, ChatGPT struggled to label images accurately at all, producing garbled characters and phantom letters regularly. Images 2.0 handles English text cleanly in his testing. The improvement is real. Where it falls apart is in languages other than English, including Chinese and Hindi, where the model's capabilities drop noticeably. That gap is a direct reflection of training data imbalance, and it represents a significant barrier for users in non-English-speaking markets.
OpenAI's rollout strategy was phased, with the core model going live December 16, 2025, for general ChatGPT users, while Business and Enterprise customers were slated for access at a later date. This staged approach is standard for high-demand launches, and it helps prevent the server overload that has plagued some previous OpenAI releases.
Key Details
- OpenAI launched ChatGPT Images 2.0 on Tuesday, December 16, 2025.
- The API version is available under the name GPT Image 1.5.
- The new model generates images up to 4 times faster than its predecessor.
- Knowledge cutoff is December 2025, more recent than most competing models.
- Aspect ratio support ranges from 3:1 wide to 1:3 tall.
- Business and Enterprise ChatGPT plan customers received delayed access after the initial rollout.
- Rogers spent several hours testing the model for Wired before publishing his hands-on assessment.
What's Next
OpenAI will need to address the non-English text rendering problem before Images 2.0 can compete meaningfully in markets like India, China, and Southeast Asia, where AI adoption is accelerating fast. Watch for OpenAI to announce targeted training improvements for at least 2 to 3 additional language groups within the next two quarters, given the competitive pressure from Google's Gemini, which has made internationalization a stated priority. The rollout to Business and Enterprise users, currently delayed, will be the next concrete milestone to track.
How This Compares
Google's Nano Banana model set the benchmark for this kind of release last year, and it did so largely because of social media virality rather than pure technical merit. The figurine trend drove millions of new users to try Google's image tools. OpenAI saw the same dynamic play out with its own caricature trend earlier this year, and Images 2.0 looks designed to create the conditions for the next wave, with multi-image outputs and faster generation giving users more to play with in a single session.
On the text rendering front, Google has also been investing heavily, as Rogers points out in his Wired piece, referencing Google's recent Nano Banana 2 improvements. Both companies have recognized that readable text inside AI-generated images is a professional use-case requirement, not a nice-to-have. OpenAI's December 2025 cutoff date arguably gives Images 2.0 a fresher knowledge base than what Google was offering at the time of the Nano Banana release, which matters for the kind of real-time infographic generation Rogers tested.
The comparison to Microsoft is also worth drawing. Microsoft has poured resources into image generation through its partnership and investment in OpenAI itself, meaning Copilot-based image tools draw from the same underlying technology stack. What differentiates Images 2.0 is the integration of reasoning capabilities directly into the image generation pipeline, which is something neither Google nor Microsoft has matched in a publicly available product as of this release. If that reasoning-plus-image approach proves sticky with professional users, it could shift the AI tools conversation away from simple prompt-to-image generation toward something more like a visual research assistant.
FAQ
Q: What is ChatGPT Images 2.0 and who can use it? A: ChatGPT Images 2.0 is OpenAI's latest image generation model, released on December 16, 2025. It is available to all ChatGPT and Codex users globally at no cost, with a more powerful version offered to paying subscribers. Developers can also access it through the API under the name GPT Image 1.5.
Q: How is Images 2.0 different from the previous ChatGPT image model? A: The new model generates images up to 4 times faster, can produce multiple images from a single prompt, connects to the internet for real-time information, and handles English text inside images far more cleanly than before. It also supports a wider range of aspect ratios, from 3:1 wide to 1:3 tall, directly from the prompt.
Q: Can ChatGPT Images 2.0 generate text in languages other than English? A: Technically yes, but not well. The model supports non-English languages like Chinese and Hindi in theory, but testing by Wired found that its performance in those languages falls significantly short of its English capabilities. This is a known limitation that OpenAI has not yet resolved.
OpenAI is iterating faster than at any point in the company's history, and ChatGPT Images 2.0 is a clear signal that visual content generation is now a core competitive battleground alongside text. The non-English gap is the one number to watch, because closing it will determine whether this model reaches global scale or stays a tool for English-first markets. Subscribe to the AI Agents Daily weekly newsletter for daily updates on AI agents, tools, and automation.
Get stories like this daily
Free briefing. Curated from 50+ sources. 5-minute read every morning.