gpt-image-2.5 sync vs batch vs Sume: cost per 1,000 images

A thousand gpt-image-2.5 images cost about $178 on OpenAI at the sync output rate, about $89 in batch, and $222.50 on Sume at the listed $0.2225 per image. The first two numbers depend on how many output tokens your images use; the Sume number is the listed per-image estimate.

The OpenAI pricing page (read 2026-10-04) lists gpt-image-2.5 and gpt-image-2 at $5 per 1M text input tokens, $8 per 1M image input tokens and $30 per 1M image output tokens, with batch halving them (image output $15).

Where the token count comes from

OpenAI bills by token, not by image, so the per-image price depends on the image. Sume's $0.2225 equals a list price of $0.178 multiplied by its 1.25 margin ($0.178 x 1.25 = $0.2225). At $30 per million output tokens, $0.178 corresponds to 0.178 / 0.00003 = about 5,933 output tokens. That token count is my inference from Sume's list figure, not an OpenAI-published number, so treat the first two rows as a worked example.

If your images use 3,000 output tokens, sync is 3,000 x $30 / 1M = $0.09, or $90 per 1,000; at 8,000 tokens it is $0.24, or $240.

What the Sume price covers

Sume's image docs list the GPT Image 2.5 model as a per-job estimate, with a reserve band that runs up to $7.36 for the largest sizes and a floor of one cent. Sume reserves at submit and captures actual usage, so a smaller image settles lower than the estimate and a failed generation is refunded.

The Sume listing has no separate batch rate, so the OpenAI batch route is cheaper at volume if you can wait for results.

• Need images now, in a pipeline with video and voice: Sume's single account and one balance.
• Overnight bulk with no deadline: batch is half price on OpenAI.

Steps

• Generate 20 images at your settings and read the output-token count from the usage field on each response.
• Average it, multiply by $30 / 1M for sync or $15 / 1M for batch.
• Compare with $0.2225 per image on Sume.
• Add a retry rate; one rerun in ten raises each total by 10%.

With a 20% retry rate

Retrying one image in five turns 1,000 keepers into 1,200 generations. At the inferred per-image prices that is 1,200 x $0.178 = $213.60 sync, 1,200 x $0.089 = $106.80 batch and 1,200 x $0.2225 = $267.00 on Sume. The retry rate scales every column equally, so it does not change which route is cheapest.

Caveats

Batch results come back asynchronously and are not meant for interactive use. Text input and image input tokens add to the OpenAI total when you send prompts or reference images. Prices change; reread both pages.