Oleksandr Kuzmenko, EditorPublished 6 min readAI-assisted · editor-reviewed
TL;DR
- Images 2.6 generates native SVG vector files, allowing direct injection into React/HTML UI components.
- Flare cuts latency to ~6 seconds per asset; Sunburst preserves structural details across sequential prompt revisions.
- API pricing mirrors previous generation: $30 per 1M output tokens, $8 per 1M reference image tokens.
Key facts
- Price per 1M image output tokens
- $30.00
- Price per 1M image input tokens
- $8.00
- Price per 1M prompt text tokens
- $5.00
- Generation speedup
- 50% faster than Images 2.5
- Flare low-quality latency
- ~6 seconds
- Global weekly volume
- 4 billion images across ChatGPT, Codex, and API (self-reported)
Source: Key facts — Price per 1M image output tokens, Price per 1M image input tokens, Price per 1M prompt text tokens
Source: Key facts. Price per 1M image output tokens: $30.00. Price per 1M image input tokens: $8.00. Price per 1M prompt text tokens: $5.00. The chart compares only values that share a unit. Each figure is copied from the story.
Metric chart
- $30.00Price per 1M image output tokens
- $8.00Price per 1M image input tokens
- $5.00Price per 1M prompt text tokens
- Price per 1M image output tokens$30.00
- Price per 1M image input tokens$8.00
- Price per 1M prompt text tokens$5.00
Direct Vector Graphics in Frontend Workflows
Historically, AI image models produced solely raster formats (PNG, JPG), requiring designers to manually trace assets in vector tools before shipping icons or logos into production. Images 2.6 introduces native SVG generation through the OpenAI API and Codex. Because SVG files store mathematical paths rather than fixed pixel grids, the generated assets scale infinitely across responsive viewport widths and can be styled dynamically via CSS fill and stroke properties.
Model Tiers: Flare vs. Sunburst
OpenAI split the release into two distinct models:
- Flare: Focuses on raw generation velocity. At low quality settings, it outputs completed assets in approximately 6 seconds, making it ideal for layout scaffolding and iterative prototyping.
- Sunburst: Targets detail retention and multi-turn consistency. When running iterative refinements, it avoids facial, text, and component drift across successive prompt turns.
Both models cost identical rates on the API: $30 per 1M output image tokens, $8 per 1M input reference tokens, and $5 per 1M prompt text tokens.
Spritesheets and Animation Controls
The update also enables GIF generation from single-line text descriptions, sketches, or reference photos. Partner integrations like OpenArt demonstrate that the model generates a 6-frame sequence arranged in a 3-column, 2-row spritesheet before compiling the final looping GIF. Additionally, the API introduces a complexity parameter, allowing developers to balance composition density against compute overhead.
Try it in 2 minutes
curl https://api.openai.com/v1/images/generations \
-H "Content-Type: application/json" \
-H "Authorization: Bearer $OPENAI_API_KEY" \
-d '{
"model": "gpt-image-2.6-flare",
"prompt": "Minimalist geometric developer terminal icon, exported as vector svg",
"response_format": "svg"
}'When to use
- When generating scalable SVGs, brand marks, and app icons directly into web components during vibe-coding sessions.
- When creating animated looping GIFs and spritesheets for product previews and marketing assets without external tooling.
When NOT to use
- Do not use for generating production UI code when your design system strictly requires pre-audited icon libraries like Lucide or FontAwesome.
- Avoid using Sunburst in tight sub-second interactive user agent loops where inference cost and latency dominate.
What to do today
- Test prompting Codex with requests for inline SVG vector icons instead of raster asset placeholders.
- Select Flare in API configurations when prototyping rapid interactive UI mockups to lower latency.
- Set the complexity parameter to high for detailed marketing landing page illustrations.
Step names are taken from What to do today.
Step names are taken from What to do today. Test prompting Codex with requests for inline SVG vector icons instead of raster asset placeholders.. Select Flare in API configurations when prototyping rapid interactive UI mockups to lower latency.. Set the complexity parameter to high for detailed marketing landing page illustrations.
- Test prompting Codex with requests for inline SVG vector icons instead of raster asset placeholders.
- Select Flare in API configurations when prototyping rapid interactive UI mockups to lower latency.
- Set the complexity parameter to high for detailed marketing landing page illustrations.