DALL·E 3 vs Flux vs Stable Diffusion: API Pricing Compared
I pulled current per-image API pricing for gpt-image-2, the retired DALL-E 3, Flux, and Stable Diffusion so you can see the real spread — and where the convenience premium actually comes from.
I get asked some version of "which image API is cheapest" a lot, and the honest answer is: it depends on the quality tier you pick, and the landscape shifted under everyone's feet this year. DALL-E 3 was pulled from the OpenAI API on May 12, 2026 — if you're comparing prices for a new project, the model you'd actually call is gpt-image-2, not DALL-E 3. I'm including DALL-E 3's last known pricing below for reference since a lot of people still search for that comparison, but treat it as historical, not something you can currently provision.
None of these numbers are locked in — providers change pricing without much notice, and quality tiers get renamed. Everything below is approximate, dated to July 2026, and worth re-checking against the provider's own pricing page before you budget a project around it.
Approximate per-image cost at 1024x1024 (mid-2026 — verify current pricing)
| Provider / Model | Tier | Approx. cost per image |
|---|---|---|
| OpenAI gpt-image-2 | Low | ~$0.006 |
| OpenAI gpt-image-2 | Medium | ~$0.053 |
| OpenAI gpt-image-2 | High | ~$0.211 |
| DALL-E 3 (retired from API, May 2026) | Standard | ~$0.04 |
| DALL-E 3 (retired from API, May 2026) | HD | ~$0.08 |
| Flux (via fal.ai) | Schnell | ~$0.003 |
| Flux (via fal.ai) | Dev | ~$0.025 |
| Flux (via fal.ai) | Pro | ~$0.04–$0.06 |
| Stable Diffusion / SDXL (via Stability AI API) | Standard | ~$0.002–$0.006 |
A few things jump out. Low-tier gpt-image-2 and Stable Diffusion are both cheap enough that per-image cost basically stops mattering unless you're generating at real volume. High-tier gpt-image-2, on the other hand, costs roughly 35x what its own low tier costs, for the same resolution — the quality slider on OpenAI's side isn't a minor toggle, it's a completely different price bracket. Flux sits in the middle and is genuinely the best-balanced option if you want better-than-low quality without paying gpt-image-2's high-tier premium.
What actually drives the price
It's not really "which model" so much as three variables stacked on top of each other:
- Quality tier. This is the single biggest lever. On gpt-image-2, moving from low to high quality is close to a 35x jump in cost for the same image size. If you're prototyping or generating thumbnails, low tier is very likely fine — save high tier for final assets.
- Resolution and aspect ratio. Square 1024x1024 is usually the cheapest option; wider or taller aspect ratios cost more because they're more output tokens or more compute, depending on the provider's billing model. OpenAI bills image generation by output tokens rather than a flat per-image fee, so the "price per image" numbers above are really a translation of a token rate into a fixed size — useful for comparison, but not literally how you're billed.
- Edits and iterations. Every one of these providers treats an edit (inpainting, variation, upscale) as its own billable generation. If your product does three rounds of "make it more X" per final image, multiply your per-image number by roughly however many rounds you actually keep.
The convenience premium, stated plainly
Here's the part I want to be upfront about: calling a model API directly, if you already know how to write the request and handle the response, is cheaper than using a hosted layer on top of it. That's just true, and I'm not going to pretend otherwise. Imagify sits on top of OpenAI's models — it doesn't replace them or beat their raw per-image price. What it removes is the setup: no API key management, no billing dashboard to configure, no code to write to get from "I have a prompt" to "I have an image." If you're comfortable writing a fetch call and handling errors and retries yourself, going direct to the provider will save you money at any real volume. If you want a prompt box and a download button today, that's the trade a hosted layer like Imagify is actually making on your behalf — convenience for a per-image markup, not a lower raw price.
If you want the fuller picture — speed, integration effort, and quality side by side rather than just price — the full API comparison covers that. And if you're trying to estimate a monthly budget rather than just a per-image number, how much AI image generation costs walks through that math.
Sources: openai.com/business/pricing, developers.openai.com/api/docs/pricing, fal.ai/pricing, platform.stability.ai/pricing, and coverage from pricepertoken.com and tokenmix.ai on 2026 API pricing changes. Verify current numbers directly with each provider before budgeting — pricing pages are the source of truth, not this post.