OpenAI Tests Next-Gen Image V2 Model on ChatGPT and LM Arena

OpenAI is quietly rolling out its next-generation image generation model, internally referred to as Image V2, through both ChatGPT and the LM Arena. This new model has surfaced in three distinct variants:

• packingtape-alpha
• maskingtape-alpha
• gaffertape-alpha

These versions have been spotted on LM Arena, a platform where AI models are often tested and compared. However, during the weekend, OpenAI pulled back images from the Arena, although they can still be encountered on ChatGPT.

Testing Framework and User Access

Some ChatGPT users have reported gaining permanent access to Image V2, while others are seeing its outputs through an A/B testing framework where they are asked to choose between competing results. This approach is familiar for OpenAI, which used a similar Arena-based blind testing method in December 2025 when it previewed models codenamed Chestnut and Hazelnut. These models eventually shipped as GPT Image 1.5 just weeks later.

Early Impressions

Early impressions of Image V2 suggest that it represents a significant step forward in image generation capabilities. Testers highlight several key improvements:

• Realistic UI Interfaces: The model can now render realistic user interface (UI) elements with correctly spelled button text, addressing a longstanding weakness in AI image generators.
• Strong Prompt Adherence: Users report that Image V2 adheres more closely to prompts, ensuring that the generated images align better with the intended output.
• Compositional Understanding: The model demonstrates improved compositional understanding, creating more coherent and contextually appropriate images.

Comparisons to Competitors

Comparisons to Google’s Nano Banana Pro are already circulating. Some users find Image V2 competitive in areas where OpenAI’s current model still lags behind Google’s offering, which has dominated the LM Arena leaderboard for months. This is particularly significant given that OpenAI has been operating under a “code red” posture since Google’s Gemini 3 and Nano Banana Pro began eroding its market position in late 2025.

Community Feedback

Several users on X (formerly Twitter) have shared their experiences with Image V2:

• Elania: Link
• Can: Link
• levelsio: Link
• Angel: Link
• Flowers: Link

Future Considerations

The key question now is whether OpenAI will maintain the model’s current quality at launch or dial it back for cost and safety reasons, a pattern the company has followed before. Pricing will also be a crucial factor. GPT Image 1.5 already undercuts its predecessor by 20% on API costs, so how Image V2 is priced could significantly impact its adoption.

Conclusion

OpenAI’s Image V2 model shows promising improvements in several key areas, potentially putting it on par with or even ahead of leading competitors like Google’s Nano Banana Pro. As the company continues to test and refine the model, practitioners will be watching closely for any updates on performance, pricing, and availability.