OpenAI has officially entered the next phase of the AI image generation race with the launch of its latest model, GPT Image 1.5. Announced on December 17, 2025, the model promises significant improvements in speed, instruction-following, and editing capabilities, aiming to challenge the current market leaders. However, initial hands-on tests and community feedback suggest that while its product integration is seamless, the raw output quality may still lag behind the established benchmark set by competitors like Google's Nano Banana Pro.
A Leap in Speed and Product Integration
The headline technical improvement for GPT Image 1.5 is its processing speed, which OpenAI claims is four times faster than its predecessor. This acceleration is designed to streamline the creative workflow, making iterative generation and editing more practical. Beyond raw speed, OpenAI has deeply integrated the model into ChatGPT, launching a dedicated image generation section. This new interface offers users a variety of pre-made templates and style options, significantly lowering the barrier to entry for creating complex or stylized images. The move underscores OpenAI's strategy of prioritizing user-friendly productization, making advanced AI tools accessible directly within a familiar chat interface.
Key Specifications & Pricing of GPT Image 1.5
- Release Date: December 17, 2025
- Claimed Speed: 4x faster than previous OpenAI image model.
- Pricing (Token-based):
- High-quality (1MP): ~USD 133 per 1,000 images
- Low-quality: ~USD 9 per 1,000 images
- Availability: Integrated into ChatGPT for all users; API released.
Performance and Pricing: A Mixed Benchmark Picture
OpenAI has adopted a token-based pricing model for GPT Image 1.5, with costs scaling based on resolution and quality settings. For high-quality, one-megapixel images, the price is approximately USD 133 per thousand generations, while low-quality images cost around USD 9 per thousand. All ChatGPT users gained access to the model upon release, with its API also becoming available for developers. In terms of official benchmarks, GPT Image 1.5 has reportedly topped both text-to-image and image editing leaderboards on the Artificial Analysis website, surpassing Google's Nano Banana Pro. It has achieved similar leading positions on the LMArena model arena, indicating strong performance in controlled testing environments.
Reported Benchmark Performance (as of Dec 17, 2025)
- Artificial Analysis: Ranked 1 in Text-to-Image and Image Editing leaderboards.
- LMArena: Ranked 1 in Text-to-Image and Image Editing leaderboards.
- Note: OpenAI has not released its own official benchmark data.
Hands-On Tests Reveal a "Viscous" Aesthetic and Detail Struggles
Despite promising benchmark scores, independent tests conducted by tech media and users reveal noticeable gaps in output quality. When generating complex scenes, such as a rainy night in Tokyo with multiple specified elements, GPT Image 1.5's results were often described as having a distinct "AI vibe" or a "viscous" feel, with oversaturated colors and unnatural blending between foreground subjects and backgrounds. Critical errors, such as generating a hand with only four fingers, were observed, which are considered basic failures for a 2025-era model. In style transfer tasks, like recreating a scene in Vincent van Gogh's The Starry Night style, the model struggled to accurately capture the distinctive brushwork and color palette of the original masterpiece, where competitors succeeded.
Community & Hands-On Test Findings vs. Nano Banana Pro
| Test Area | GPT Image 1.5 Observations | Nano Banana Pro Observations (for comparison) |
|---|---|---|
| Aesthetic | Often "viscous," oversaturated, has an "AI vibe." | More natural, photographic quality; minor imperfections add realism. |
| Detail Accuracy | Struggles with complex prompts (e.g., missing hand details). Prone to errors like incorrect anatomy. | High accuracy in replicating prompt details. |
| Style Transfer | Can miss core artistic elements (e.g., Van Gogh's brushwork). | More faithful to the requested style's characteristics. |
| Edge Cases | Performance degrades significantly (e.g., distorted perspectives). | Handles unusual prompts more robustly. |
| Speed | Faster than its predecessor, but reportedly slower than Nano Banana Pro in some tests. | Noted for fast generation times (~15 seconds in one test). |
| Strength | Strong instruction-following for edits, good multi-element fusion, excellent product integration. | High baseline of realism and detail consistency. |
Edge Cases and Realism Lag Behind the Competition
The model's weaknesses become more pronounced in edge-case scenarios. A prompt requesting a first-person perspective from a cat resulted in a severely distorted image with inconsistent details, failing to convey the requested viewpoint convincingly. In side-by-side comparisons shared online, portraits generated by GPT Image 1.5 often exhibited proportion issues, such as oversized heads, and less natural lighting compared to those from Nano Banana Pro. Users noted that Nano Banana Pro's outputs sometimes included minor imperfections like slightly overexposed windows, which ironically contributed to a greater sense of photographic realism. Some testers found that GPT Image 1.5's results could be improved by adding specific photographic directives like "unprocessed iPhone photo" to the prompt, suggesting its default aesthetic is overly processed.
Strengths in Editing and Multi-Element Fusion
Where GPT Image 1.5 shows considerable promise is in its image editing and compositing capabilities. OpenAI demonstrated the model's ability to perform complex element swaps—changing clothing colors, vehicle types, and street signs within an existing image with high accuracy. It also supports multi-element fusion, such as convincingly placing multiple people and a dog into a single cohesive scene based on a descriptive prompt. The built-in style templates allow for rapid transformations, like turning a corporate headshot into an 80s VHS fitness tape cover or a scene into a pink, early-2000s doll-game aesthetic, showcasing its utility for quick, creative remixing.
Conclusion: A Polished Product Awaiting a Model Breakthrough
The launch of GPT Image 1.5 represents a solid, iterative update from OpenAI, excelling in user experience, speed, and creative control through editing and templates. Its seamless integration into ChatGPT makes powerful image generation more accessible than ever. However, the consensus from early testing is that the core model still trails the current state-of-the-art in critical areas like realism, detail consistency, and reliability with complex or unusual prompts. In a market where user expectations have been sharply raised by rivals, GPT Image 1.5 delivers a superior product experience but leaves room for improvement in the underlying generative engine. Its market reception will ultimately depend on how much users value streamlined workflow over ultimate output fidelity.