GPT Image 2 — OpenAI's next image model, tested as "ducttape" on LMArena
GPT Image 2 (GPT-Image-2) is the successor to OpenAI's GPT-Image-1. It has been quietly testing on LMArena under the codename ducttape since late 2025 — and the community has been watching closely. Here is everything known so far about capabilities, release timeline, and how it compares to the competition.
What is GPT Image 2?
GPT Image 2 is OpenAI's next-generation image generation model. While GPT-Image-1 brought native image output to ChatGPT and the OpenAI API, GPT Image 2 represents a significant leap forward — particularly in text rendering, layout composition, and multilingual support.
The model first appeared on LMArena in late 2025 under the anonymous codename "ducttape", sometimes listed as duct-tape 1, duct-tape 2, and duct-tape 3 — variants that represent different model versions or parameter presets. Among these, duct-tape 2 has consistently produced the most balanced output across prompt types.
GPT Image 2 capabilities
Based on extensive community testing on LMArena, GPT Image 2 excels in three areas that have historically been weak spots for image generation models:
Text rendering
Posters, packaging, menus, signage, album covers, and UI mockups come out with legible, properly kerned typography. Multi-line text stays on baseline. Spelling errors are rare.
Layout fidelity
When a prompt specifies a composition — "three panels, the third one vertical" — GPT Image 2 obeys it. Reference images are preserved with accurate angle, crop, and palette.
Multilingual output
Korean, Japanese, and Chinese characters render correctly — not romanised, not mangled. This is still rare among image models as of early 2026.
GPT Image 2 examples from LMArena
Every image below was generated by GPT Image 2 (ducttape) during blind testing on LMArena. No post-processing, no cherry-picking.
Get notified the moment GPT Image 2 launches publicly.
GPT Image 2 vs the competition
On LMArena, GPT Image 2 runs head-to-head against the best image models from Google, ByteDance, and Black Forest Labs. Here is how it stacks up based on community blind-vote results:
GPT Image 2 vs Nano Banana Pro (Gemini 2.5 Flash Image)
Nano Banana Pro leads on photoreal portraits and face consistency. GPT Image 2 wins on in-image text, multilingual characters, and explicit layout instructions.
GPT Image 2 vs Seedream 4 (ByteDance Doubao)
Seedream 4 leads on stylised illustration, manga aesthetics, and East-Asian design references. GPT Image 2 wins on multi-panel compositions and reference-image fidelity.
GPT Image 2 vs Flux 1.1 Pro (Black Forest Labs)
Flux leads on open-weight workflows, ultra-realistic micro-detail, and LoRA fine-tunes. GPT Image 2 wins on poster text, UI mockups, and packaging copy.
GPT Image 2 vs GPT-Image-1
GPT Image 2 is a major generational leap over GPT-Image-1 in text rendering accuracy, layout structure, and multilingual support. It consistently outperforms GPT-Image-1 in LMArena blind votes.
GPT Image 2 release date
No official release date has been announced by OpenAI. GPT Image 2 has been testing on LMArena under the ducttape codename since late 2025, with the most recent activity in April 2026.
For context: in mid-2025 the codename "nanobanana" appeared on LMArena and was confirmed as Google's Gemini 2.5 Flash Image within roughly six to eight weeks. GPT Image 2 has been in testing longer than that, suggesting a public launch could be imminent.
When GPT Image 2 launches — whether as a ChatGPT feature, an API endpoint, or both — subscribers to this site will be the first to know.
GPT Image 2 API
GPT-Image-1 is available through the OpenAI API with support for text-to-image and image editing. GPT Image 2 is expected to follow the same pattern, likely accessible via the gpt-image-2 model identifier.
No API endpoint, pricing, rate limits, or resolution options have been announced yet. Once the API goes live, this page will be updated with integration guides, pricing comparisons, and code examples.
GPT Image 2 FAQ
What is GPT Image 2?
GPT Image 2 (GPT-Image-2) is OpenAI's next-generation image generation model, the successor to GPT-Image-1. It has been testing on LMArena under the codename 'ducttape' since late 2025.
Is ducttape GPT Image 2?
Yes. The codename 'ducttape' on LMArena has been confirmed to be GPT Image 2. The model appeared as duct-tape 1, duct-tape 2, and duct-tape 3 — representing different model versions.
When will GPT Image 2 be released?
No official date has been announced. The model has been in LMArena testing since late 2025, and based on precedent (nanobanana → Gemini 2.5 Flash Image in 6-8 weeks), a release could come soon.
How is GPT Image 2 different from DALL-E?
GPT Image 2 is part of the GPT-Image line, which is separate from DALL-E. While DALL-E is a standalone model, GPT-Image models are natively integrated into ChatGPT and the OpenAI API. GPT Image 2 significantly outperforms both DALL-E 3 and GPT-Image-1.
Can I use GPT Image 2 now?
Not directly. The only way to experience GPT Image 2 currently is through LMArena's blind A/B testing interface, where you compare two anonymous model outputs. You cannot select it by name.