What an AI Image Actually Costs in 2026: 14 Chinese Models Compared

Quick facts (checked 2026-10-22): 111 live models through one OpenAI-compatible API · 0% platform markup, provider list price passed through · free tier gives 10 free-model calls/day · machine-readable prices: china-ai-leaderboard.json · agent index: llms.txt

Image pricing is quoted in two incompatible ways: a flat price per image, or a token price where each image burns a different number of tokens depending on its size. This table converts both into one generated image, so the numbers can be compared directly.

Cost per image

Model	Provider	Billing	Per image
CogView3-Flash	Zhipu AI	free	FREE
Wan 2.2 T2I Flash	Alibaba Cloud	flat	$0.0215
Seedream 4.0	ByteDance	flat	$0.0270
Seedream 5.0	ByteDance	flat	$0.0270
Wan 2.5 T2I Preview	Alibaba Cloud	flat	$0.0308
Hunyuan Image 3.0	Tencent	token (20K)	~$0.0308
Seedream 4.5	ByteDance	flat	$0.0340
Seedream 5.0 Pro	ByteDance	flat	$0.0340
Seedream v5.0 Lite (TokenHub)	ByteDance	token (22K)	~$0.0338
WAND Vega Image Lite	Tencent	token (22.5K)	~$0.0346
Seedream v5.0 Pro (TokenHub)	ByteDance	token (60K)	~$0.0923
Vidu Image Q2	Vidu	token (93.75K)	~$0.1442
WAND Vega Image Flash	Tencent	token (100.8K)	~$0.1551
WAND Vega Image Pro	Tencent	token (171K)	~$0.2631

How the token rows are computed: token price ($1.5385 per 1M) x the model's estimated tokens per image. Those rows are estimates; the flat-price rows are the published per-image rates.

What the spread is actually buying

Resolution and detail tier. WAND Vega Lite and Pro are the same family: 4.5x more tokens per image at the Pro tier.
Model generation. Seedream 4.0, 4.5 and 5.0 sit within 25% of each other — newer is not automatically dearer.
Token vs flat billing. A flat $0.027 image (Seedream 5.0) and a token-billed ~$0.031 image (Hunyuan Image 3.0) land in the same place; the billing model matters less than the effective price.

At scale

Volume	Cheapest paid	Seedream 5.0	WAND Vega Pro
100 images	$2.15	$2.70	$26.31
1,000 images	$21.50	$27.00	$263.10
10,000 images	$215	$270	$2,631

For product shots and social assets, Seedream-class models at $0.027 cover most needs. The 10x jump to WAND Vega Pro is only worth it when the image is the product — hero images, print, or assets that will be retouched anyway.

Free tier for pipeline testing

CogView3-Flash costs nothing on NovAI: 10 free-model calls/day and 5 requests/minute on the free tier. It is enough to wire up generation, storage and CDN delivery before spending on the good models.

curl https://aiapi-pro.com/v1/images/generations \
  -H "Authorization: Bearer $NOVAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"cogview-3-flash","prompt":"a paper boat in the rain, macro"}'

FAQ

Why do some image models cost 10x more per image?

Three factors: output resolution and quality tier (lite vs pro), the provider's compute cost, and the billing model — some models charge a flat price per image, others bill per token and the image consumes 20K-171K tokens depending on its size.

Is there a free image model?

Yes. CogView3-Flash is free to call with a NovAI API key; the free tier allows 10 free-model calls per day and 5 requests per minute.

Are these prices NovAI's or the providers'?

They are the providers' list prices passed through with a 1.0 multiplier and zero platform markup, published at /china-ai-leaderboard.json so they can be checked.

How much does 1,000 images cost?

At the cheapest paid rate (Wan 2.2 T2I Flash, $0.0215/image) about $21.50; with Seedream 5.0 at $0.027 it is $27; with WAND Vega Pro at $0.263 it is $263. The free tier covers testing only.

Every image model, one endpoint

111 live models, provider list price, 0% platform markup. $2 credit on signup, no card required.

Get an API key

Prices reflect provider rate cards as of 2026-10-22 and are published machine-readably at /china-ai-leaderboard.json. Token-billed figures are estimates based on each model's tokens-per-image; flat figures are the published rates.