What OpenAI’s image line is
OpenAI, the makers of ChatGPT, also build image models. KADR has five: GPT-5 Image, GPT-5 Image Mini, GPT-5.4 Image 2 and, since October 2026, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst. Flare costs 2, 5 and 19 tokens for low, medium and high quality, Sunburst costs 3, 6 and 25.
The line’s overall character: it understands complex descriptions better than most competitors. If you have a long prompt with details, conditions and logic («the man on the left holds a red book, there are three objects on the table, rain outside the window»), OpenAI’s models parse it more precisely. They also handle text inside the image more confidently.
Five versions and their prices
| Version | Price per image | Feature |
|---|---|---|
| GPT-5.4 Image 2 | 2 / 6 / 17 tokens (~3 / 9 / 27 rubles) | low / medium / high quality |
| GPT Image 2.5 Flare | 2 / 5 / 19 tokens (~3 / 8 / 30 rubles) | low / medium / high quality |
| GPT Image 2.5 Sunburst | 3 / 6 / 25 tokens (~5 / 9 / 40 rubles) | low / medium / high quality |
| GPT-5 Image Mini | 7 tokens (~11 rubles) | budget, simple tasks |
| GPT-5 Image | 29 tokens (~46 rubles) | flagship, maximum quality |
This is not a public offer. KADR prices, October 2026.
The GPT-5.4 Image 2 trick: pay by quality
This model (like the two GPT Image 2.5 models) has something the older versions do not: three quality levels at different prices. Roughly (KADR data):
- low about 2 tokens,
- medium about 6 tokens,
- high about 17 tokens.
It is convenient. You literally pay for the level you need: drafts and tests on low for pennies, the final shot on high. No need to overpay for quality that does not matter at the selection stage.
Where OpenAI’s line beats the competition
The three main image models have different roles:
- GPT-5 Image is understanding complex descriptions and text.
- Flux.2 is pure photorealism from scratch.
- Nano Banana Pro is editing existing photos.
The practical takeaway: if the prompt is long and precision matters (exactly these objects, in exactly this arrangement, with this text), take OpenAI’s models. If you just need a nice, simpler realistic picture, Flux is cheaper and will cope. If you need to change an existing photo, that is Nano Banana.
How to choose a version
GPT-5.4 Image 2 is the most flexible option: thanks to quality-based pricing you pay exactly as much as you need. Good as a workhorse.
GPT-5 Image Mini is for simple tasks where complex understanding is not needed but cheaper volume is.
GPT-5 Image is the flagship for complex, important shots where maximum prompt adherence matters. But it is the most expensive image model in the line (46 rubles), so do not burn it on drafts.
About text in the image
OpenAI’s models handle text inside an image reasonably well, better than average. But Cyrillic, as with everyone, is weaker than Latin script. For guaranteed readable Russian text it is more reliable to generate the image without a caption and add the text in an editor, or to use Nano Banana Pro, which handles short Cyrillic captions cleanly. More in the article on product cards, where this problem is covered.
Who it suits
Suits: complex scenes that must follow a long description precisely, images with text, tasks where compositional logic matters.
Not the main choice: simple photorealism (Flux is cheaper) and edits to existing photos (Nano Banana is better).
In short
OpenAI’s image line is strong at understanding complex prompts and text. Five versions: GPT-5.4 Image 2 with flexible pay-by-quality pricing, the newer GPT Image 2.5 Flare and Sunburst priced the same way, Mini for simple tasks, GPT-5 Image as the expensive flagship. Take them when precision and details in the prompt matter, and for pure photorealism or editing there are more specialised and cheaper models.
