OpenAI ChatGPT Images 2.0 Adds Reasoning and Web Search
OpenAI rolled out ChatGPT Images 2.0, its updated image generator based on the GPT Image 2 model. This version includes reasoning steps before creation and web search capabilities. Users can now generate as many as eight consistent images from one prompt. The tool also manages text much better overall, with strong results for non-Latin scripts.
The company confirmed the release in a blog post. GPT Image 2 mirrors features in Google's Nano Banana Pro by taking time to think based on the chosen mode. It adjusts reasoning duration and pulls web data when needed. OpenAI expects this to boost image variety and precision. Paid ChatGPT Plus, Pro, and Business subscribers alone get the full thinking outputs.
Thinking Mode Enables Multi-Image Consistency
When users turn on thinking mode, ChatGPT Images 2.0 produces up to eight images simultaneously from a single instruction. Characters, objects, and styles remain uniform across outputs. OpenAI cites examples like full-page mangas from one image and text prompt, sets of social media visuals, and house room design series.
Quality Upgrades Reach Every User
All ChatGPT accounts benefit from enhanced image quality, no matter the mode. The generator now grasps key photo traits more accurately. It excels in pixel art, manga, film stills, and similar styles. Previous models faltered on details like tiny text, icons, UI components, crowded scenes, and nuanced style cues. This one handles them well.
Supported aspect ratios stretch from 3:1 ultra-wide to 1:3 ultra-tall. These fit banners, slides, and phone displays. API access reaches 2K resolution.
API Details and Token-Based Costs
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Developers access the model as gpt-image-2 through OpenAI's API. Charges follow a token system: $8 per million image input tokens, $30 per million image output tokens. Text inputs run $5 per million, outputs $10 per million. Cached inputs cost less.
Actual per-image prices depend on quality and size. OpenAI's pricing page lists a 1024 x 1024 low-quality image at $0.006, medium at $0.053, high at $0.211. A 1024 x 1536 version costs $0.005 low, $0.041 medium, $0.165 high. Other sizes like 1536 x 1024 match those rates.
| Model | Quality | 1024 x 1024 | 1024 x 1536 | 1536 x 1024 | |, , , , , , -|, , , , -|, , , , , , -|, , , , , , -|, , , , , , -| | GPT Image 2 | Low | $0.006 | $0.005 | $0.005 | | | Medium | $0.053 | $0.041 | $0.041 | | | High | $0.211 | $0.165 | $0.165 | | GPT Image 1.5 | Low | $0.009 | $0.013 | $0.013 | | | Medium | $0.034 | $0.05 | $0.05 | | | High | $0.133 | $0.2 | $0.2 |
GPT Image 2 undercuts prior versions at bigger sizes. High-quality 1024 x 1536 runs $0.165 against $0.20 for GPT Image 1.5 and $0.25 for GPT Image 1.5 at other large formats. Standard 1024 x 1024 high quality costs more at $0.211 versus $0.133. Outputs over 2K stay in beta with possible inconsistencies.
OpenAI points to uses in localized ads, infographics, school materials, design apps, and creative sites. Codex integrates image generation right in the workspace, no extra API key required.
Real-World Tests and Pre-Launch Buzz
Tests show ChatGPT Images 2.0 performs well on tough prompts. One benchmark asked for a hyper-realistic DSLR photo: a monkey holding a pink banana sits on a tiger upfront. Behind, a horse rides an astronaut, who serves as a living spacesuit saddle. The horse controls as rider, fully clear, no reversal. High-res, sharp focus, real lighting.
Instant mode yields a somewhat artificial result. Thinking mode captures true DSLR quality better.
Before launch, codenamed gpt-image-2 reached select US testers. Samples on X and Reddit looked photo-real. It shines on complex scenes, diagrams, text-heavy screenshots, suiting ads and education like infographics. OpenAI fixed the smooth skin and ideal lighting flaws from GPT Image 1.5, where Nano Banana Pro led. A livestream at 12 pm PT revealed it. Examples included a fake Nadella image touting Chrome downloads via Edge and a teased AI screenshot.
Related on Neura Market