Fast everyday generation
Same token rates as Sunburst. Tuned for speed: lower latency on standard text-to-image work. Pick it when you want the image quickly.
OpenAI bills GPT Image 2.5 by the token, not by the image — so the real per-image price depends on quality and resolution. Here are the official rates, measured costs, and the cheapest way to use it.
Three honest paths: developers pay the raw API — roughly $0.013–$0.053 per 1K image depending on quality. ChatGPT subscribers already have it with no per-image billing. Everyone else is better off with an all-in-one tool: ModelPixLab plans work out to $0.025–$0.049 per image, and new accounts get 3 free images with no card. The raw API is the cheapest per image — tools charge for not having to manage keys, code, and billing.
Official rates
There is no flat per-image price. Both gpt-image-2.5-flare and gpt-image-2.5-sunburst bill per million tokens at identical rates:
| Rate (per 1M tokens) | What counts | |
|---|---|---|
| Text input | $5.00 | Your prompt words |
| Image input | $8.00 | Reference/edit images you attach |
| Image output | $30.00 | The generated image — this dominates the bill |
Published OpenAI API rates as documented across pricing references in 2026. OpenAI can change rates or token accounting at any time — check their pricing page before budgeting a project.
Per-image math
Token counts below come from an independent measurement (own API key, September 2026) at 1024×1024. Your prompt length and reference images move the number.
| Output tokens | Approx. cost | |
|---|---|---|
| Medium quality, 1K | ~439 | ~$0.013 |
| High quality, 1K | ~1,756 | ~$0.053 |
| 2K / 4K, high quality | Higher token counts | $0.10+ (gateway list prices: ~$0.10 at 2K, ~$0.16 at 4K) |
Measured values are a snapshot, not a guarantee. Editing with reference images adds input-image tokens on top.
Flare vs Sunburst
A common question — the answer is simple:
Same token rates as Sunburst. Tuned for speed: lower latency on standard text-to-image work. Pick it when you want the image quickly.
Same token rates as Flare. Tuned for tighter control on edits and intricate detail. Pick it when you are iterating on an existing image.
Price is not a reason to choose between them — task fit is.
Cheapest route
Honest answer: the raw API wins on per-image price, but price isn't the whole cost.
New ModelPixLab accounts get 3 free images, no credit card. Enough to judge the model before spending anything.
~$0.013 per 1K image is the cheapest measured route. You manage the key, the code, retries, and billing — fine if that's already your stack.
If you pay for ChatGPT, generation inside ChatGPT has no per-image charge — it falls under your subscription's usage limits instead.
No keys, no token math, editing UI included. Compare per-image equivalents below — pick the tier that matches your volume.
On ModelPixLab
Our plans converted to a per-image equivalent (1 credit = 1 generation), straight from the pricing page:
| Price | Images | Per image | |
|---|---|---|---|
| Free trial | $0 | 3 (one-time) | $0 |
| Starter pack | $4.9 one-time | 100 | $0.049 |
| Creator pack | $12.9 one-time | 300 | $0.043 |
| Studio pack | $29.9 one-time | 800 | $0.037 |
| Basic | $9.9/mo | 250/mo | $0.040 |
| Pro | $19.9/mo | 700/mo | $0.028 |
| Studio | $39.9/mo | 1,600/mo | $0.025 |
Credit packs never expire; subscriptions renew monthly and cancel anytime. Heavy users land near $0.025/image with zero API setup.
FAQ
3 free images, no credit card. See what the model does before you think about pricing at all.
Also see: GPT Image 2.5 vs Midjourney · GPT Image 2.5 vs Nano Banana 2 · prompt guide