GPT Image 2.5 Batch is $15 vs $30 per M output tokens; Sume has none

OpenAI's pricing page lists GPT Image 2.5 image output at $30.00 per million tokens on Standard and $15.00 on Batch (read 2026-10-09), so Batch halves the output cost. Sume's Image API docs describe no batch tier for the same model, so on Sume you pay the catalog row for each image.

OpenAI's rates

The page lists both gpt-image-2.5-sunburst and gpt-image-2.5-flare at the same rates. Every line is exactly half on Batch.

What that is per image

Sume's docs publish an output-token estimate for a 1024 x 1024 image: at max quality the output is $0.21072 at $30 per million tokens, which is 7,024 tokens (0.21072 / 0.00003). At Batch's $15 per million the same tokens cost $0.10536. For 1,000 such images the output alone is $210.72 on Standard against $105.36 on Batch, before any input tokens.

Those figures are output-only and exclude Sume's margin and input tokens, so they are not what Sume bills. On Sume, GPT Image 2.5 is billed from catalog rows: low 1K $0.02475, medium 2K $0.055625 and high 4K $0.2225.

Choosing

Direct Batch is the cheaper route for large, non-urgent GPT Image 2.5 volume at high quality. Sume suits work where you want the same request shape across models and results in seconds. A middle route is to lower the quality tier: Sume's low row ($0.02475) is below even the Standard-rate max-quality output estimate ($0.21072) by about 8.5 times, and quality is the bigger cost lever than the batch discount.

Remember that on OpenAI cached-input rates apply only to images generated with the Responses API, per the pricing page note.

What to check

Before you move volume to Batch, confirm three things on OpenAI's side: which API surface the Batch rate applies to, how long the batch window is, and whether your workflow can wait for it. None of those is described in the rates table above, and this post does not claim them.

On the Sume side, check GET /v1/images/models for the GPT Image 2.5 descriptors you need: quality levels, input_references (up to 16) and background. The two GPT Image 2.5 ids on Sume, openai/gpt-image-2.5 and openai/gpt-image-2.5-sunburst, share the same rates.