An image generator’s cost is not always a flat fee per finished picture. GPT Image 2.5 quotes text and image meters separately, so resolution, iteration and rejected renders affect the budget.
Pricing snapshot: Vendor first-party documentation checked on 25 October 2026. Original price effective date unknown unless separately declared. Amounts are published list figures, subject to tier, region, currency and eligibility.
The model family and dated identifiers
The image API reference includes gpt-image-2.5-sunburst and gpt-image-2.5-flare, along with dated identifiers ending in 2026-09-08. Record the actual ID used. These are not the same thing as the retired DALL-E model IDs or the earlier gpt-image-2. The reference also notes specific shutdown dates for older models; a historical article must not project a replacement’s features or pricing backward in time.
Prices are quoted by the provider in USD per million text/image tokens, not an unconditional price per generated image. A public marketing page may describe example image equivalents, but the authoritative charge for an individual request depends on metering and settings.
Understand the rate card
The Standard pricing table lists $8 per million image-input tokens, $30 per million image-output tokens and $5 per million text-input tokens for GPT Image 2.5 Sunburst and Flare. Cached image input is listed at $2, with cached text input at $1.25 per million. The Batch table shows separate, lower prices, including $15 per million image output. Choosing a Batch row should happen only when the actual API job used that processing path.
Do not describe $30 per million output image tokens as “$30 per image.” Equally, do not assume every output at a chosen resolution consumes a fixed published token count without checking the vendor’s image cost calculator and response usage.
Estimate a campaign from accepted assets
A commercial workflow might generate three candidate images per approved scene, edit one of them twice, then upscale or composite the selected output elsewhere. A useful budget counts every generation and editing operation, including discarded attempts, rather than only the number of final assets. Define acceptance criteria—composition, legible embedded text, brand safety and licensing checks—before forecasting the number of iterations.
For a token-metered hypothetical job with 500,000 billed image-output tokens and 30,000 text-input tokens at Standard rates, the model components would be $15 + $0.15. That is an illustration of the formula, not a prediction of how many pictures those token counts yield.
Separate rights, quality and spend
A low price does not prove ownership of reference material or permission to reproduce a person’s likeness. The generation service’s commercial terms, user-provided image rights and any required disclosures need legal review. Technical teams should budget for manual quality assurance where hallucinated labels, hands, products or factual scenes matter, rather than presenting a generated image as authenticated evidence.
For image-to-image workflows, budget input image tokens as well as generated output. Changes to quality or aspect ratio may change the bill; a fixed-price spreadsheet that omits them is brittle.
When to use a per-image comparison
Per-image comparisons can be useful only after normalizing model, size, quality, route, editing passes, and what constitutes one accepted deliverable. For APIs with a published per-image schedule, use that schedule; for token-metered GPT Image, derive a per-accepted-image cost from recorded usage. Do not transplant rates from Midjourney subscriptions or Adobe credit plans into this token ledger.
The checked vendor pages do not announce when every listed rate first took effect; effective_at remains null. Keep the price record and rights review distinct. Batch results, resolution changes and product availability should be rechecked before purchase.
Pricing evidence boundary
This cost discussion concerns OpenAI (gpt-image-2.5-sunburst; gpt-image-2.5-flare). Reproduce a quote only with its exact model or plan identifier, chargeable unit, billing region, processing mode and timestamp; a changing vendor rate card is not a permanent price guarantee. If any one of these dimensions is missing, keep the associated rate and worked scenario under editorial review instead of filling the gap with an estimate.
Sources & Further Reading
- developers.openai.com — pricing (official documentation; accessed 2026-10-25)
- developers.openai.com — generate (official documentation; accessed 2026-10-25)