GPT Image 2.6 (Flare and Sunburst): GIF, SVG, pricing and what you can use from Russia

• What’s new. Images 2.6 in ChatGPT and Codex on every plan, plus the gpt-image-2.6-flare and gpt-image-2.6-sunburst models in the API.
• Flare vs Sunburst. Flare is the fast model for everyday generation (OpenAI says that with quality: "low" an image can be ready in as little as about six seconds). Sunburst is for detailed work: complex compositions and consistent details across edits.
• New in the API. A complexity parameter, GIF output (output_format: "gif") and SVG output (output_format: "svg"). GIF and SVG are generation-only, not available for edits.
• OpenAI’s price. The same rates as GPT Image 2 and 2.5: $5 per million text input tokens, $8 for image input, $30 for image output. The Batch API is not supported for 2.6.
• On RuAPI. The 2.6 models aren’t available yet. gpt-image-2.5-flare, gpt-image-2.5-sunburst and gpt-image-2 are — from Russia, without a VPN or a foreign card.

​

What OpenAI announced

• animated GIFs from a prompt or a sketch, with adjustable playback speed; the feature is rolling out gradually;
• saved reference images (yourself, pets, characters) that you can reuse in later chats without uploading them again.

​

Flare vs Sunburst

​

What’s new in the API

​

The complexity parameter

​

Animated GIFs

• Request: output_format: "gif" on POST /v1/images/generations.
• Frame count via frame_count (default 9); frame duration via frame_duration_ms (default 200 ms; the guide gives 20–2000 ms, rounded to 10 ms).
• The frame_count limit differs between docs: the API reference says 2 to 16, the image generation guide says up to 64. Check what the API accepts on your account.
• No streaming, no partial images (partial_images), no compression setting and no prompt_rewriter.
• The API reference says GIF output requires access to the GIF preview for the selected model, so it may not be enabled on every account at first.

​

Vector SVG

• Request: output_format: "svg" on POST /v1/images/generations, n=1 only.
• The SVG is written by a separate “author” model configured in the svg object: svg.model and svg.reasoning.effort (from none to max). Per the API reference, this requires access to SVG model selection; the reference doesn’t list the allowed models.
• With SVG output, complexity controls the intermediate raster reference image, not the SVG author’s reasoning.
• No streaming, partial images, compression or prompt_rewriter; the edit endpoint neither returns SVG nor accepts it as input.

​

Markup-guided edits

​

GPT Image 2.5 vs 2.6

​

Pricing

​

How to use it from Russia: GPT Image 2.5 on RuAPI

curl -sS https://www.ruapi.ai/v1/images/generations \
  -H "Authorization: Bearer sk-YOUR_KEY" \
  -H "Content-Type: application/json" \
  -o resp.json \
  -d '{
    "model": "gpt-image-2.5-flare",
    "prompt": "A minimalist coffee shop logo: a cup with steam shaped like a cloud",
    "size": "1024x1024",
    "quality": "medium",
    "n": 1
  }'

tr -d '\\' < resp.json | grep -o '"b64_json": *"[^"]*"' | head -n 1 | cut -d'"' -f4 | base64 --decode > image.png
[ -s image.png ] && echo "Saved: image.png" || cat resp.json

import base64

from openai import OpenAI

client = OpenAI(api_key="sk-YOUR_KEY", base_url="https://www.ruapi.ai/v1")

result = client.images.generate(
    model="gpt-image-2.5-flare",  # or gpt-image-2.5-sunburst
    prompt="A minimalist coffee shop logo: a cup with steam shaped like a cloud",
    size="1024x1024",
    quality="medium",
    n=1,
)

with open("image.png", "wb") as f:
    f.write(base64.b64decode(result.data[0].b64_json))

• model — gpt-image-2.5-flare for fast generation, gpt-image-2.5-sunburst when you need precision.
• quality — low, medium or high; it directly affects the number of output tokens, i.e. the price.
• size — e.g. 1024x1024, 1536x1024, 1024x1536.

​

GPT Image vs Nano Banana

• GPT Image is called via /v1/images/generations and /v1/images/edits, returns the image in b64_json, and is billed per token.
• Nano Banana is called via /v1/chat/completions, like a regular chat, and returns the image inside the response text; some models have a fixed per-image price, others are billed per token.

​

FAQ