OpenAI's GPT Image models are billed per output token, not per image — which makes the real cost hard to see. Here's the true cost per 1024×1024 image, derived from OpenAI's token pricing, for every GPT Image model and quality tier.
Get our weekly newsletter on pricing changes, new releases, and tools.
| Model | Low | Medium | High |
|---|---|---|---|
| GPT Image 1.5 (current)gpt-image-1.5 | $0.009 | $0.034 | $0.133 |
| GPT Image 1 Mini (cheapest)gpt-image-1-mini | $0.005 | $0.011 | $0.036 |
| GPT Image 1 (legacy — retires Oct 2026)gpt-image-1 | $0.011 | $0.042 | $0.167 |
Per-image prices are for a single 1024×1024 output image; portrait/landscape (1024×1536) cost ~1.5×. Prompt text tokens and any input images are billed on top.
Unlike most image APIs, OpenAI doesn't charge a flat per-image price — it meters output image tokens. A 1024×1024 image is a fixed token count per quality tier, so the per-image cost is simply tokens × the output-token rate:
So on gpt-image-1.5 ($32 / 1M output tokens), a high-quality image is 4,160 × $32/1M = $0.133. That's the number you actually pay — and it's what we show, so GPT Image lines up against per-image models like Flux and Imagen on the cheapest image API ranking.
gpt-image-1.5 is the current flagship and the best quality. gpt-image-1-mini is dramatically cheaper (from ~$0.005/image) for high-volume or draft work. gpt-image-1 is the original model and is being retired on Oct 23, 2026 — avoid building new on it. For most use cases, medium quality on 1.5 (~$0.034/image) is the sweet spot; switch to mini when cost dominates.
Built by @aellman
Also from the founder: BlobCap, a price index for AI training data deals, showing what AI labs actually pay for data.
2026 68 Ventures, LLC. All rights reserved.