1,000 drafts: GPT Image 2.5 low at $10 vs Seedream 5.0 Lite at $50
A thousand draft images cost $10 on GPT Image 2.5 at quality low and $50 on Seedream 5.0 Lite on Sume. The cents arithmetic and when the dearer row still wins.
A thousand draft images cost $10 on ChatGPT Image 2.5 at quality: "low" (1,000 x 1 cent) and $50 on Seedream 5.0 Lite (1,000 x 5 cents) on Sume. So the cheaper draft row is not the one with a lite name. A cheaper still option is Qwen Image at 3 cents, which is $30 per thousand and the cheapest row that also accepts reference images.
The arithmetic
Prices below come from the Sume image catalog (list price times 1.25, rounded up to the cent) and the GPT Image 2.5 quality tiers; they are as of 2026-10-08.
| Model | Billed per image | Per 1,000 |
|---|---|---|
| ChatGPT Image 2.5, low | 1 cent | $10 |
| ChatGPT Image 2.5, medium | 2 cents | $20 |
| Qwen Image | 3 cents | $30 |
| Seedream 5.0 Lite | 5 cents | $50 |
| ChatGPT Image 2.5, high | 7 cents | $70 |
Why quality tier matters more than model name
It is natural to assume a 'Lite' model is the economy choice. In the catalog, Seedream 5.0 Lite lists at $0.035, which becomes $0.04375 billed and 5 cents with rounding, five times the low tier of GPT Image 2.5. The tier of the call is what moves the bill.
But low quality is not the same output as lite. If the drafts are used only to pick a composition, the 1-cent tier is the right place to start. If the drafts will be shown to a customer, medium at 2 cents is still under Seedream's price.
- Drafts for internal review: GPT Image 2.5 low.
- Customer-facing options: GPT Image 2.5 medium (2 cents) or a row you have tested.
- Edits from a photo at volume: Qwen Image (3 cents).
- Final art: GPT Image 2.5 high (7 cents), or the row that wins your test.
Where the dearer row can still win
Seedream 5.0 Lite lists nine aspect ratios including 4:5 and 3:2, and GPT Image 2.5 lacks 3:2. If your layout needs a ratio the cheap row does not list, the price comparison stops mattering. The same goes for a style you prefer: run 20 prompts on each, which costs 20 cents and $1.00, and look.
Remember the n field: n: 4 on any row multiplies the price by four, since Sume bills cost_usd x n, so a 1,000-image job is 250 calls at n=4.
Budget guard
Read usage.cost from each response and stop when the running sum passes your cap. A simple loop over 1,000 prompts at low quality should end near $10; if it does not, check whether you omitted quality, because the default is high and a thousand high images are $70.
Rounding, once more
The cents above are the catalog's rounded-up figures. The unrounded Seedream figure is $0.04375, so 1,000 images come to $43.75 if the wallet is charged unrounded and $50 at the catalog's cents. The docs say usage.cost is the amount billed, so log it for a pilot of 50 images and multiply to see which figure your account follows.
Sources
Related posts
More in Pricing
- 1,000 transparent PNGs: gpt-image-2.5 background vs model plus RMBG
A transparent PNG costs $0.0074 to $0.066 from gpt-image-2.5 alone, or the model price plus $0.0225 RMBG for other rows. 1,000-image table on Sume.
- 1080p premium by row: Recast +50%, H3 Max +100%, H3 has no 1080p
Moving up to 1080p adds 50% to H3 Max Recast and 100% to MiniMax H3 Max on Sume, while MiniMax H3 stops at 768p. Per-second and per-10-second figures.
- 11-minute clip, 10-minute transcript cap: two passes cost $0.15
Sume transcripts cover up to 10 audio minutes. Split an 11-minute clip with two $0.02 trims, then 10 + 1 transcript minutes at $0.01 each: $0.15.
- 11-second H3 Max clip: $0.69 at 480p, $1.10 at 768p, $2.20 at 1080p
An 11-second MiniMax H3 Max clip with native stereo audio costs $0.69 at 480p, $1.10 at 768p and $2.20 at 1080p on Sume. MiniMax H3 prices included.
Written by Sume