GPT-Image-2.5 Is Live — Flare and Sunburst at 50% Off, Same Token Rates as GPT-Image-2

OpenAI released ChatGPT Images 2.5 on September 8, 2026, calling it their new state-of-the-art image model. Unlike previous image launches, this one arrives in the API as two models with different speed/precision trade-offs. We enabled both on Smart AIPI the same day.

Below: what changed, how to pick between flare and sunburst, why the price did not move, and working code.

What Changed in Images 2.5

OpenAI describes four concrete improvements over the previous generation:

• More natural lighting and richer textures. The most visible difference on photorealistic prompts — surfaces read as material rather than as rendered gradients.
• Better subject preservation from reference photos. A face, a product, or a logo you pass in as a reference survives the generation with fewer identity drifts.
• More reliable multi-turn editing. Editing instructions are followed more consistently across several turns of a conversation, which is the single biggest failure mode of iterative image workflows.
• Up to 50% lower latency than Images 2.0. Same request, roughly half the wait.

One note for completeness: OpenAI also shipped two ChatGPT-only features alongside this release — Sketch (draw directly in ChatGPT as a reference) and Templates (starting points for flyers, product photos and similar formats). Neither is part of the image API, so neither is available through Smart AIPI or any other API provider.

Flare or Sunburst? A Straight Answer

The two models are priced identically, so the choice is purely about latency versus control.

A practical pattern: explore with flare at quality: "low" until the prompt is right, then render the final asset once with sunburst at quality: "high". You pay draft prices for the iteration and premium quality only for the frame you keep.

Pricing: Unchanged, and Still 50% Off

OpenAI kept the token rates identical to GPT-Image-2 for both 2.5 models:

• Text input — $5.00 per 1M tokens ($1.25 cached)
• Image input — $8.00 per 1M tokens ($2.00 cached)
• Image output — $30.00 per 1M tokens

That means the speed and quality upgrade costs developers nothing extra, and our per-image prices are exactly what they were for gpt-image-2 — 50% of OpenAI retail:

Non-square sizes at or above 1024×1536 bill at 1.5× the square rate — so a high-quality 1536×1024 render is $0.1583 on Smart AIPI. Your $5 in free signup credits therefore still covers roughly 1,650 low-quality drafts, 188 medium renders, or 47 high-quality hero images — now generated by a better model, in about half the time.

Generate: Basic Request

curl https://api.smartaipi.com/v1/images/generations \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-image-2.5-flare",
    "prompt": "A photorealistic ceramic mug on a weathered oak table, morning light",
    "size": "1536x1024",
    "quality": "high"
  }'

The response is the standard OpenAI shape — { "created": ..., "data": [{ "b64_json": "..." }] }. Base64-decode b64_json to get the image bytes.

Python SDK

import base64
from openai import OpenAI

client = OpenAI(base_url="https://api.smartaipi.com/v1", api_key="YOUR_API_KEY")

result = client.images.generate(
    model="gpt-image-2.5-sunburst",
    prompt="Studio product shot of a matte black espresso machine on seamless grey",
    size="2048x2048",
    quality="high",
)

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

Editing: Multi-Reference and Masks

Both 2.5 models accept multiple reference images in a single edit call, and mask-based inpainting where white regions are rewritten and black regions are preserved. This is where the improved multi-turn instruction following shows up most clearly — a sequence of edits drifts away from your intent less than it did on gpt-image-2.

curl https://api.smartaipi.com/v1/images/edits \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=gpt-image-2.5-sunburst" \
  -F "image[][email protected]" \
  -F "image[][email protected]" \
  -F "image[][email protected]" \
  -F "prompt=Photorealistic gift basket with a ribbon containing the referenced items"

curl https://api.smartaipi.com/v1/images/edits \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "model=gpt-image-2.5-flare" \
  -F "image=@sunlit_lounge.png" \
  -F "[email protected]" \
  -F "prompt=Replace the pool with a flamingo float"

The mask must match the primary image's pixel dimensions, be a PNG with an alpha channel, and stay under 50MB.

Model IDs and Aliases on Smart AIPI

• gpt-image-2.5-flare — the new default. Faster, and OpenAI's recommendation for most applications.
• gpt-image-2.5-sunburst — precision and premium creative. Slower.
• gpt-image-2.5 — alias for flare.
• gpt-image-2, gpt-image-1.5, gpt-image-1, gpt-image-1-mini — still available by explicit model ID.
• dall-e, dall-e-3, gpt-image, gpt-image-latest, or an empty model field — all route to gpt-image-2.5-flare.

Sizes, Formats and Parameters

• size — auto or any WxH where both edges are multiples of 16, max edge ≤ 3840px, total pixels between 655,360 and 8,294,400, and the long:short ratio is ≤ 3:1. The same free-form geometry gpt-image-2 introduced, including 4K.
• quality — Smart AIPI bills and supports low, medium, high and auto (defaults to medium). OpenAI's 2.5 models additionally expose xhigh and max tiers upstream.
• output_format — png (default), jpeg, webp. Pair jpeg/webp with output_compression (0–100).
• background — auto or opaque. Transparent is not supported on this family and returns HTTP 400; use gpt-image-1.5 for an alpha channel.
• n — batch size. Smart AIPI loops server-side because the underlying image tool produces one image per call.
• image / mask (edits only) — base64 string, array of base64 strings, or multipart file parts.

Get Started

1. Sign up at smartaipi.com/signup (free credits, no credit card)
2. Create an API key in the dashboard
3. Set your base URL to https://api.smartaipi.com/v1
4. Use model gpt-image-2.5-flare — or send nothing and get it by default

Sources: OpenAI's Introducing ChatGPT Images 2.5 announcement and the gpt-image-2.5-sunburst model reference.