Cheapest Image Generation API 2026: 10 Models per Image

The cheapest image generation API on Orux AI as of Oct 29, 2026 is Nano Banana 2.1 at 3.6 Credits ($0.018) per 1K image, which is 360 Credits ($1.80) per 100 images. GPT Image 2.5 Flare and Sunburst come next at 5.4 Credits ($0.027), with Wan2.7 Image effectively tied at 5.42 Credits. At 4K the order changes: Seedream 5.0 is 8.07 Credits per image against 8.1 for Nano Banana 2.1.

Below are all ten image models on Orux AI with per-image prices at every resolution, the cost per 100 images, the official prices we could verify, and the one model where Orux AI charges more than the vendor does.

Price per image, ranked

Orux AI prices as of Oct 29, 2026, sorted by the lowest available resolution. 1 Credit = $0.005; dollar figures are that conversion.

Swipe horizontally to view the full table.

Notes:

• OpenAI prices GPT Image 2.5 by tokens, not per image: $30 per 1M image output tokens, $8 per 1M image input tokens and $5 per 1M text input tokens. There is no official flat per-image price to compare against. Orux AI charges a fixed price per image by resolution, so you know the cost before you send the request.
• Seedream prices are published by Volcano Engine in CNY. We list them as published rather than converting at a rate that will drift.
• Seedream 5.0 and Seedream 4.5 have no 1K tier on Orux AI; their cheapest output is 2K. Ranked by cheapest available image, they sit in the middle. Ranked at 2K, they move up.

Where Orux AI costs more

Qwen Image 3.0 Pro is more expensive on Orux AI than from Alibaba Cloud directly: 10.42 Credits ($0.052) versus $0.04 at 1K (130%), and 20.83 Credits ($0.104) versus $0.075 at 2K (139%). If you specifically need Qwen Image 3.0 Pro at volume, buying it direct is cheaper. If you only need Qwen's text-to-image, standard Qwen Image 3.0 on Orux AI is exactly at the official $0.03.

Everything else in the table with a verifiable official price is at or below it: Nano Banana 2.1 at 54% of Google's 1K price, Wan2.7 Image and Wan2.7 Image Pro at 90%.

Resolution changes the ranking

Most comparison tables quote the cheapest tier and stop. If you ship 2K or 4K images, the order is different.

Swipe horizontally to view the full table.

What stands out:

• At 2K, Nano Banana 2.1 (5.4) and Wan2.7 Image (5.42) are tied for cheapest, and GPT Image 2.5 drops to fourth at 9 Credits.
• At 4K, Seedream 5.0 (8.07) and Nano Banana 2.1 (8.1) are effectively tied, and GPT Image 2.5 costs 14.4, nearly 1.8 times as much.
• Nano Banana 2.1's 4K price is 36% of Google's official $0.113, the biggest discount in this table.
• Seedream 5.0 Pro and Qwen Image 3.0 Pro both double from 1K to 2K. Wan2.7 and the two flat-priced Seedream models do not charge more for higher resolution at all.

Per 100 images at 2K: Nano Banana 2.1 540 Credits ($2.70), Wan2.7 Image 541.7 ($2.71), Seedream 5.0 806.7 ($4.03), GPT Image 2.5 900 ($4.50), Seedream 4.5 958 ($4.79), Wan2.7 Image Pro 1,354.2 ($6.77), Qwen Image 3.0 Pro 2,083.4 ($10.42), Seedream 5.0 Pro 2,100 ($10.50).

Speed

Generation time matters if a user is waiting. Orux AI shows an average time per completed task on each model page; the samples are small, so treat these as rough indicators, not benchmarks.

Swipe horizontally to view the full table.

The other six models do not have enough completed tasks on Orux AI to report an average. Flare and Sunburst cost the same per image, so if latency matters and both fit your style, Flare's lower average is a free win.

Image editing and reference images

None of these prices change based on whether the output contains text. What differs is how many reference images each model accepts and whether that costs extra. These are parameter facts from each model's request schema on Orux AI.

Swipe horizontally to view the full table.

If you need editing or multi-image composition on a budget, Nano Banana 2.1 is the cheapest model that takes reference images. If you need the most references in one request, GPT Image 2.5 takes up to 16. For lettering, each model page has an existing Orux example whose prompt asks for the words "Orux AI" or "ORUX CAFÉ", but we have not scored text accuracy across models, so check your own prompts.

For deeper coverage of each model, see the Nano Banana 2.1 API guide, GPT Image 2.5 API pricing, Seedream 5.0 Pro API pricing and the Qwen Image 3.0 API guide.

How to choose

A simple way to pick, using only the numbers above:

• High-volume drafts, thumbnails or placeholder art at 1K: Nano Banana 2.1 at 3.6 Credits. Nothing else on Orux AI is cheaper per image.
• Production images at 2K: Nano Banana 2.1 or Wan2.7 Image at about 5.4 Credits. Wan2.7 Image charges the same at 1K and 2K, so there is no reason to render it at 1K.
• 4K output: Seedream 5.0 or Nano Banana 2.1 at about 8.1 Credits. GPT Image 2.5 at 4K costs 14.4.
• Many reference images in one request: GPT Image 2.5 (up to 16) or Seedream 5.0 and 4.5 (up to 14, no surcharge).
• Lowest latency in our small sample: GPT Image 2.5 Flare at 67 seconds on average.

Run two candidates on ten of your real prompts before committing. A model that needs two attempts per usable image has doubled its effective price, and that matters more than a 1-Credit difference in list price.

One script, any image model

All ten models use the same async task endpoint on Orux AI: submit to POST /api/v1/tasks, poll GET /api/v1/tasks/{task_id}, download the result. Only the input fields differ. This script runs one prompt through any model you name and prints the list-price cost.

# pip install requests
import os, sys, time, requests

BASE = "https://orux.top/api/v1"
HEADERS = {"Authorization": f"Bearer {os.environ['ORUX_API_KEY']}"}

# Credits per image at the resolution used below, checked 2026-10-29
MODELS = {
    "nano-banana-2.1":        ("1K", 3.6),
    "gpt-image-2.5-flare":    ("1K", 5.4),
    "wan2.7-image":           ("1K", 5.42),
    "qwen-image-3.0":         ("1K", 6.0),
    "seedream-5.0":           ("2K", 8.07),  # no 1K tier
    "seedream-5.0-pro":       ("1K", 10.5),
}

def build_input(model, prompt, resolution):
    inp = {"prompt": prompt, "resolution": resolution}
    # Qwen Image 3.0 only supports 1:1; the rest accept 16:9
    inp["aspect_ratio"] = "1:1" if model == "qwen-image-3.0" else "16:9"
    return inp

def generate(model, prompt):
    resolution, credits = MODELS[model]
    r = requests.post(f"{BASE}/tasks", headers=HEADERS, timeout=60, json={
        "model": model, "input": build_input(model, prompt, resolution)})
    if r.status_code == 402:
        raise SystemExit("Insufficient balance: top up at https://orux.top/pricing")
    r.raise_for_status()
    task_id = r.json()["task_id"]

    while True:
        time.sleep(5)
        t = requests.get(f"{BASE}/tasks/{task_id}", headers=HEADERS, timeout=30).json()
        if t["status"] not in ("QUEUED", "RUNNING"):
            break
    if t["status"] != "SUCCESS":
        print(f"{model}: {t['status']} {t.get('error')} (not charged)")
        return
    for i, url in enumerate(t["result"]):
        path = f"{model}_{i}.png"
        open(path, "wb").write(requests.get(url, timeout=120).content)  # URLs are temporary
        print(f"{model}: saved {path}, {credits} Credits at list price")

if __name__ == "__main__":
    prompt = "A ceramic mug on a wooden desk, soft morning light, no text"
    for name in sys.argv[1:] or ["nano-banana-2.1"]:
        generate(name, prompt)

Run python images.py nano-banana-2.1 gpt-image-2.5-flare wan2.7-image to compare three models on the same prompt for 14.42 Credits ($0.072) at list price. Image tasks have no callback option on Orux AI, so polling is the pattern. More request details are in /docs/tasks.

Prices checked on 2026-10-29. Sources: Gemini API pricing, OpenAI pricing, Alibaba Cloud Model Studio pricing, Volcano Engine Ark pricing (CNY), and the Orux AI model pages under /models?modality=image.

FAQ

Q.Which image generation API is cheapest?

On Orux AI as of Oct 29, 2026, Nano Banana 2.1 is cheapest at 3.6 Credits ($0.018) per 1K image, or 360 Credits ($1.80) per 100 images. At 2K it ties with Wan2.7 Image at about 5.4 Credits, and at 4K Seedream 5.0 is marginally cheaper at 8.07 Credits.

Q.Does 4K cost more than 1K?

It depends on the model. Nano Banana 2.1 goes from 3.6 to 8.1 Credits and GPT Image 2.5 from 5.4 to 14.4 Credits. Seedream 5.0 and Seedream 4.5 charge the same at 2K and 4K, and Wan2.7 Image charges the same at 1K and 2K.

Q.Which image APIs support image editing?

Every model here except standard Qwen Image 3.0 accepts reference images through image_urls. GPT Image 2.5 takes up to 16, Seedream 5.0 and 4.5 up to 14, Nano Banana 2.1 and Seedream 5.0 Pro up to 10, Wan2.7 up to 9, and Qwen Image 3.0 Pro up to 3.

Q.Are failed image generations billed?

No. On Orux AI, image tasks are charged only when they reach SUCCESS. Tasks that end as FAILED or EXPIRED cost nothing. You pay from a prepaid balance with no monthly fee, so a rejected prompt does not eat into your budget.

Q.Is Orux AI always cheaper than the official API?

No. Nano Banana 2.1 is 36–54% of Google's official price and Wan2.7 is 90% of Alibaba Cloud's, but Qwen Image 3.0 Pro costs 30–39% more than buying from Alibaba Cloud directly. Check the "Orux vs official" column before you commit volume to a model.