OpenAI just pulled the trigger on GPT Image 1.5 three weeks early—and if you’re wondering why, look no further than Sam Altman’s November “code red” memo. After Google’s Nano Banana Pro gained 13 million users in four days and Gemini’s user base jumped from 450 million to 650 million between July and October 2025, OpenAI accelerated its January 2026 image model release to December 16, 2025. The message is clear: this isn’t innovation, it’s defense.
But defensive doesn’t mean weak. GPT Image 1.5 delivers 4x faster generation, 35% better instruction adherence over DALL-E 3, and tops both Artificial Analysis leaderboards with 1,260 ELO for text-to-image and 1,270 ELO for editing. It’s now available to all ChatGPT users (Free, Go, Plus, Edu, Pro) and via API at 20% lower pricing than GPT Image 1. The model also introduces something genuinely useful: localized editing that preserves everything you don’t touch.
What is GPT Image 1.5?
GPT Image 1.5 is OpenAI’s latest image generation model, released December 16, 2025. Unlike DALL-E 3, which operated as a separate system, GPT Image 1.5 is natively multimodal—built on the GPT-5.2 reasoning architecture. This means it understands images and text as part of its core structure, not through separate processing pipelines.
The model supports three resolutions: 1024×1024 (square), 1536×1024 (landscape), and 1024×1536 (portrait). It’s accessible through the ChatGPT interface for all users and via API under the name gpt-image-1.5. Business and Enterprise users will get access to the new dedicated “Images” workspace later, though the model itself is already live.
The performance breakthrough: 4x faster, #1 on benchmarks
Let’s start with the numbers. GPT Image 1.5 generates images up to 4x faster than GPT Image 1, which launched in April 2025. That’s not incremental improvement—that’s the difference between iterating through concepts quickly and waiting around long enough to lose your creative momentum.
On the Artificial Analysis leaderboards, GPT Image 1.5 currently holds the #1 position with 1,260 ELO for text-to-image generation and 1,270 ELO for image editing. That puts it ahead of Google’s Nano Banana Pro (Gemini 3 Pro Image), which previously dominated both categories. The model also shows a 35% improvement in instruction adherence compared to DALL-E 3, meaning it’s significantly better at actually creating what you ask for.
OpenAI reports significant improvements in dense text rendering—the ability to generate readable text within images, which matters for creating infographics, presentation slides, and marketing materials. Multi-step sequential editing also works more reliably, letting you refine images through multiple iterations without quality degradation.
The editing revolution: localized control without the chaos
Here’s where GPT Image 1.5 gets genuinely interesting. The model introduces precise localized editing that preserves everything you don’t explicitly change. Modify a specific area—say, changing someone’s expression or adjusting an object’s position—and the model maintains facial likeness, lighting, composition, and color tone across the rest of the image.
This matters more than it sounds. Previous image models often treated edits like regeneration, introducing subtle changes throughout the entire image even when you only wanted to modify one element. GPT Image 1.5’s approach means you can iteratively refine images without the “drift” that used to accumulate across multiple edits.
The model also supports concurrent output generation, letting you generate multiple variations simultaneously for A/B testing. Combined with the 4x speed improvement, this makes GPT Image 1.5 significantly more practical for production workflows where you need to compare options quickly.
Pricing: 20% cheaper than GPT Image 1
OpenAI reduced API pricing by 20% compared to GPT Image 1. While exact pricing varies by resolution and quality settings, estimates suggest approximately $0.01 for low quality, $0.04 for medium quality, and $0.17 for high quality images. ChatGPT users get access through their existing subscriptions (Free, Go, Plus, Edu, Pro) with no additional charges.
The pricing reduction combined with 4x faster generation means GPT Image 1.5 offers roughly 5x better cost-efficiency per unit time compared to its predecessor. For high-volume users running production workflows, that translates to meaningful savings.
The code red context: Google’s Gemini momentum
To understand why OpenAI accelerated this release, you need context. On November 18, 2025, Google launched Gemini 3, which immediately claimed first place on LMArena’s AI leaderboard. Gemini 3 beat GPT-5.1 on AIME 2025 math tests (96.7% vs 89%), Humanity’s Last Exam (41% vs 26.5%), and multimodal reasoning benchmarks.
More concerning for OpenAI: Gemini’s user base grew from 450 million monthly active users in July to 650 million in October 2025. Google’s Nano Banana Pro (Gemini 3 Pro Image) went viral following its release, gaining 13 million users in just four days. The model’s photorealistic output and natural lighting earned praise for “passing vibe checks” despite GPT Image 1 leading on technical benchmarks.
Sam Altman’s internal “code red” memo put every other OpenAI project on the backburner to focus on ChatGPT improvements. Image generation was specifically mentioned as a key priority. GPT Image 1.5 was originally planned for January 2026—the three-week acceleration tells you how seriously OpenAI takes the Gemini threat.
What GPT Image 1.5 still can’t do
OpenAI acknowledges several limitations. The model has limited support for certain drawing styles, particularly artistic techniques that require specific aesthetic interpretation. Scientific knowledge accuracy sits around 70%, meaning you shouldn’t trust it for technical diagrams without verification.
GPT Image 1.5 struggles with multiple small faces in a single image, often producing distorted or inconsistent facial features when rendering crowds or groups. Non-English text rendering needs improvement—the model handles English text significantly better than other languages, which limits its usefulness for international markets.
These aren’t minor limitations. If your use case involves scientific visualization, crowd scenes, or multilingual content, you’ll run into frustrating edge cases. OpenAI is transparent about these issues, which beats discovering them in production.
GPT Image 1.5 vs Nano Banana Pro: speed versus vibe
GPT Image 1.5 leads technical benchmarks, but user feedback suggests a more nuanced picture. Nano Banana Pro (Google’s Gemini 3 Pro Image) produces more photorealistic results with better natural lighting and an “unedited” aesthetic that many users prefer. GPT Image 1.5’s outputs sometimes feel slightly processed or artificial by comparison.
For text rendering, complex edits, and fast iteration, GPT Image 1.5 is the clear winner. It generates faster, costs less, and handles sequential modifications better. For pure photorealism and that “just took this with my phone” vibe, Nano Banana Pro still has the edge.
The competitive positioning matters: GPT Image 1.5 optimizes for production workflows and practical utility. Nano Banana Pro optimizes for aesthetic quality and natural look. Choose based on whether you prioritize speed and editing flexibility or photorealistic output.
The new ChatGPT Images interface
OpenAI redesigned ChatGPT’s image generation experience with a dedicated “Images” tab in the sidebar. The interface includes preset styles, trending prompts, and templates designed to reduce the friction of getting started. It’s rolling out now for Free, Go, Plus, Edu, and Pro users, with Business and Enterprise access coming later.
The interface overhaul suggests OpenAI is betting on image generation as a core feature, not a side project. Combined with the model improvements, it’s a clear signal that visual AI is central to OpenAI’s strategy for competing with Google’s momentum.
A defensive move that still delivers
GPT Image 1.5 is fundamentally a defensive release—accelerated to counter Google’s Gemini momentum rather than launched on its original timeline. But defensive doesn’t mean insignificant. The model delivers meaningful improvements in speed, cost-efficiency, and editing capabilities that make it genuinely more useful for production workflows.
The code red context matters because it reveals OpenAI’s strategic position. Three years ago, Google declared its own code red after ChatGPT’s launch. Now the roles are reversed, and OpenAI is scrambling to match Google’s growth and benchmark performance. GPT Image 1.5 is one move in a longer game—and based on Anthropic’s Claude Code release and other recent developments, the game is far from over.
For users, the takeaway is straightforward: GPT Image 1.5 is fast, flexible, and practical. If you need to generate and iterate on images quickly—especially with complex text or sequential edits—it’s now the benchmark leader. Just don’t expect it to win every vibe check.
Get the Daily Pulse
Sharp analysis on what's actually moving in AI. No hype, no filler, no weekly digest.
