GPT-Image-2.6: Faster Creation and Editing

​

Key Highlights

• Two models are live: APIYI offers gpt-image-2.6-flare and gpt-image-2.6-sunburst through the standard /v1/images/generations and /v1/images/edits endpoints
• 50% faster generation: OpenAI reports a 50% speed improvement for Images 2.6, with Flare generating in as little as six seconds at low quality
• Clear roles: Flare targets rapid iteration and better layouts, while Sunburst emphasizes detail and edit consistency at medium and high quality
• New upstream creative controls: GIF generation, SVG export, image markup, and a complexity setting were announced with Images 2.6
• Unchanged pricing: Text input, image input, and image output remain $5 / $8 / $30 per 1M tokens, with the same cached-input rates as 2.5

​

Background

​

Deep Dive

​

Choosing between the two models

​

Four new upstream capabilities

• GIF generation: Create looping GIFs from a prompt for brand mascots, reactions, memes, and animated storyboards.
• Image markup: Mark specific areas of an image so editing instructions can target the intended region more precisely.
• SVG export: Carry designs into websites, apps, and print workflows without a separate bitmap-to-vector conversion step.
• complexity control (currently unsupported): This upstream setting controls scene-planning effort, but APIYI currently ignores it silently. Requests do not fail, but the field has no effect. Do not send or rely on it.

​

Practical Use

​

Code samples

import base64
import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["APIYI_API_KEY"],
    base_url="https://api.apiyi.com/v1"
)

# Flare: quickly draft a product poster
flare = client.images.generate(
    model="gpt-image-2.6-flare",
    prompt="A minimalist specialty coffee poster with morning light and space reserved for a headline",
    size="1536x1024",
    quality="low"
)
with open("coffee-flare.png", "wb") as f:
    f.write(base64.b64decode(flare.data[0].b64_json))

# Sunburst: refine the scene while preserving product and brand details
with open("coffee-product.png", "rb") as image:
    sunburst = client.images.edit(
        model="gpt-image-2.6-sunburst",
        image=image,
        prompt="Keep the packaging, logo, and text unchanged; use a warm wooden coffee-shop table as the background",
        size="1024x1024",
        quality="high"
    )
with open("coffee-sunburst.png", "wb") as f:
    f.write(base64.b64decode(sunburst.data[0].b64_json))

​

Best practices

1. Draft before refining: Explore compositions with Flare at low quality, then use Sunburst for final assets to reduce iteration time.
2. State what must not change: Include constraints such as “keep the packaging, logo, and text unchanged” to reduce detail drift during edits.
3. Measure cost from responses: Equal token rates do not guarantee equal per-image costs. Record input and output token usage in production.
4. Validate new fields separately: Test GIF, SVG and image markup independently before adoption. Do not send complexity; it is currently ignored silently and has no effect.

​

Pricing and Availability

​

Pricing

​

Available groups

​

Stack recharge promotions

​

Summary and Recommendations

• Everyday generation, poster drafts, and batch assets: choose gpt-image-2.6-flare
• Brand assets, complex compositions, and multi-turn retouching: choose gpt-image-2.6-sunburst
• Production integrations: verify existing generation and editing first, then adopt new capabilities individually and evaluate actual token usage and latency