AI Image Generation Cost Analysis

I've benchmarked 33 image generation models from 7 providers to compare cost and latency using the same prompt:

A photorealistic image of a golden retriever puppy sitting in a sunflower field at golden hour, with a soft bokeh background and warm light. Aspect ratio: 1:1, size: 1024x1024.

The benchmark includes both image-only and image-capable multimodal models available on Vercel AI Gateway.

If you don't have much time to analyze the full report:

• go straight to the winners
• compare images
• check out the cost chart
• or the latency chart

Full Results

July 10, 2026: added google/gemini-3.1-flash-lite-image, bfl/flux-2-max, recraft/recraft-v2, openai/gpt-image-1.5, openai/gpt-image-1-mini, and ByteDance Seedream models (seedream-4.0, seedream-4.5, seedream-5.0-lite); renamed google/gemini-3.1-flash-image-preview to its GA slug google/gemini-3.1-flash-image; prodia/flux-fast-schnell generates again. Full changelog

(click on a thumbnail to see the full image)

Model	Price	Latency	Image
google/gemini-2.5-flash-image	$0.0390494	6.9s
google/gemini-3.1-flash-image	$0.067024	11.5s
google/gemini-3.1-flash-lite-image	$0.034012	3.6s
google/gemini-3-pro-image	$0.137304	29.9s
google/imagen-4.0-fast-generate-001	$0.02	5.7s
google/imagen-4.0-generate-001	$0.04	8.0s
google/imagen-4.0-ultra-generate-001	$0.06	11.0s
bfl/flux-2-flex	$0.05	15.3s
bfl/flux-2-klein-4b	$0.014	4.2s
bfl/flux-2-klein-9b	$0.015	2.9s
bfl/flux-2-pro	$0.03	8.5s
bfl/flux-2-max	$0.07	25.4s
bfl/flux-pro-1.1	$0.04	4.3s
bfl/flux-pro-1.1-ultra	$0.06	6.9s
bfl/flux-kontext-pro	$0.04	6.9s
bfl/flux-kontext-max	$0.08	11.5s
recraft/recraft-v2	$0.022	4.6s
recraft/recraft-v3	$0.04	6.6s
recraft/recraft-v4	$0.04	14.0s
recraft/recraft-v4-pro	$0.25	17.2s	(2048x2048)
recraft/recraft-v4.1	$0.035	8.7s
recraft/recraft-v4.1-pro	$0.21	13.8s	(2048x2048)
recraft/recraft-v4.1-utility	$0.035	10.2s
recraft/recraft-v4.1-utility-pro	$0.21	18.6s	(2048x2048)
openai/gpt-image-1	$0.16667	42.7s
openai/gpt-image-2	$0.0237	37.4s
openai/gpt-image-1.5	$0.045454	32.1s
openai/gpt-image-1-mini	$0.033388	30.7s
xai/grok-imagine-image	$0.02	6.2s
bytedance/seedream-4.0	$0.03	9.4s	(2048x2048)
bytedance/seedream-4.5	$0.04	22.9s	(2048x2048)
bytedance/seedream-5.0-lite	$0.035	30.3s	(2048x2048)
prodia/flux-fast-schnell	$0.0025	8.5s

Total spent: $2.0351014

The latency here is wall time, measured by the benchmark script. The cost, however, is returned by the gateway, so it should be accurate.

Cost

Latency

Model Quirks

This benchmark includes provider-specific adaptations, because several models need custom params to work correctly:

• Black Forest Labs Flux models need explicit pixel width and height instead of relying on aspectRatio
• Recraft and OpenAI image models require size rather than aspectRatio
• recraft/recraft-v4-pro needs 2048x2048 for square output, not 1024x1024
• xAI Grok models support aspectRatio but not size
• Gemini image responses can contain duplicate files in result.files, so the benchmark deduplicates them before saving
• ByteDance Seedream models require size (not aspectRatio) at a minimum of 3.69 MP, so they run at 2048x2048, and take watermark: false to drop the "AI generated" badge they otherwise burn into the image
• prodia/flux-fast-schnell, the cheapest model, generates again after failing in the previous run - the earlier failure looks like a transient provider issue.

Winners

If you only optimize for cost, the three cheapest models in this benchmark are:

• prodia/flux-fast-schnell at $0.0025
• bfl/flux-2-klein-4b at $0.014
• bfl/flux-2-klein-9b at $0.015

If you optimize for latency, the three fastest models are:

• bfl/flux-2-klein-9b at 2.9s
• google/gemini-3.1-flash-lite-image at 3.6s
• bfl/flux-2-klein-4b at 4.2s

I personally think it all depends on your task and prompt. For my particular mass-scale case, I'm optimizing for both cost and speed, so choose among the winners.

DIY

If you want to run the benchmark yourself, add more models, or adjust the prompt - the script is easy to configure and run: kometolabs/ai-image-generation-cost-analysis

Support My Work

Preparing such research sometimes takes 2-5 full benchmark runs, which adds up. If it saved you time, consider sponsoring me on GitHub or Buy Me a Coffee.