If you found the GPT Image 1.5 API in a tutorial or saw it listed as a model option, the honest answer about where it stands fits in a sentence: it is OpenAI’s previous image generation model, deprecated with a December 1, 2026 shutdown date, and it has exactly one reason to exist over its successor gpt-image-2 — transparent backgrounds. If you need alpha-channel PNGs from the GPT Image family, 1.5 is your only in-family option for as long as it is live. For everything else, gpt-image-2 is the better and longer-lived choice.
This page covers the model’s status, its trade-off against gpt-image-2, what it costs per resolution and quality tier, rate limits, working code, and when you should actually reach for it. If you want the full GPT Image family overview and the two API access paths, start with the OpenAI image generation API guide. If you need the endpoint reference with full parameter tables and error codes, that is covered there as well — this page stays focused on the 1.5 model itself. Every fact below is checked against OpenAI’s own documentation as of July 30, 2026.
One disclosure: we run Modellix, an aggregator that offers GPT Image 1.5 alongside 210+ other image and video models through a single REST API. We have a commercial interest when Modellix appears. All model IDs, parameters, and prices below were pulled from OpenAI’s and Modellix’s own documentation and verified –no-cache on July 30, 2026.
What gpt-image-1.5 is — and its shutdown clock
OpenAI describes gpt-image-1.5 plainly as “our previous image generation model, with better instruction following and adherence to prompts.” The alias gpt-image-1.5 points at the snapshot gpt-image-1.5-2025-12-16, and it accepts text and image input, returning image and text output. [src: https://developers.openai.com/api/docs/models/gpt-image-1.5]
The number that should drive your decision is the shutdown date. gpt-image-1.5 was deprecated in the June 2, 2026 batch of GPT Image retirements, alongside gpt-image-1-mini and chatgpt-image-latest — all three go dark on December 1, 2026, and all three point to gpt-image-2 as the replacement. (gpt-image-1 runs slightly longer, to October 23, 2026.) [src: https://developers.openai.com/api/docs/deprecations] So from today, gpt-image-1.5 has roughly four months of guaranteed life. That does not make it useless — but any integration you build on it is a countdown, not a foundation.
There is one oddity worth knowing. If you read the API reference’s type definitions, the image model enum is literally ImageModel = "gpt-image-1.5" or "dall-e-2" or "dall-e-3" or 2 more — gpt-image-1.5 sits at the front, and gpt-image-2 is not in that list at all. Meanwhile the models catalog calls gpt-image-2 the “State-of-the-art image generation model” and does not list 1.5. [src: https://developers.openai.com/api/docs/models] Do not let the stale enum fool you into thinking gpt-image-1.5 is the newest thing — it is the outgoing one. For the full lineup and deprecation schedule across all GPT Image models, see the OpenAI image generation API guide.
gpt-image-1.5 vs gpt-image-2: the trade-off that matters
For most work, gpt-image-2 is simply the better model — it is newer, it is not deprecated, and it handles up to 4K (3840×2160) resolution. For a full breakdown of all four GPT Image 2 endpoints including the Responses API path, see the GPT Image 2 API guide. There is exactly one place it loses to gpt-image-1.5, and that place is the reason this article exists.
gpt-image-2 does not support transparent backgrounds. OpenAI’s guide is explicit: “gpt-image-2 doesn’t currently support transparent backgrounds. Requests with background: "transparent" aren’t supported for this model.” [src: https://developers.openai.com/api/docs/guides/image-generation] gpt-image-1.5 supports transparency natively — so if you need a cut-out PNG with an alpha channel for UI icons, standalone logos, game sprites, or product shots on no background, gpt-image-1.5 is your only in-family option for as long as it is live.
Two related capabilities live on the same side of the split. gpt-image-1.5’s edit variant can merge up to 16 reference images into one output and exposes a fidelity control for how strictly the model preserves the input’s shape. gpt-image-2’s edit endpoint accepts a single input image, processes it at high fidelity automatically, and does not offer transparency or the strict-fidelity knob. [src: https://developers.openai.com/api/docs/guides/image-generation]
The cost of picking 1.5 is resolution: its maximum output is 1536×1024, with no 2K or 4K tier. So the trade-off table is short:
| Need | Model |
|---|---|
| Transparent background / alpha PNG | gpt-image-1.5 |
| Merge multiple reference images (up to 16) | gpt-image-1.5 (edit) |
| Strict shape preservation (fidelity control) | gpt-image-1.5 (edit) |
| 2K / 4K resolution | gpt-image-2 |
| Highest fidelity, photorealism, longevity | gpt-image-2 |
| Anything else, especially new builds | gpt-image-2 |
What gpt-image-1.5 costs
gpt-image-1.5 is priced per image, by quality tier and size. From OpenAI’s own model page:
| Quality | 1024×1024 | 1024×1536 | 1536×1024 |
|---|---|---|---|
| Low | $0.009 | $0.013 | $0.013 |
| Medium | $0.034 | $0.050 | $0.050 |
| High | $0.133 | $0.200 | $0.200 |
[src: https://developers.openai.com/api/docs/models/gpt-image-1.5]
A useful sanity check: gpt-image-1.5 generates images by first producing specialized image tokens with fixed counts per size and quality — a high-quality 1024×1024 image is 4,160 image output tokens. Image output tokens on gpt-image-1.5 cost $32 per million. [src: https://developers.openai.com/api/docs/models/gpt-image-1.5] Multiply: 4,160 × $32 / 1,000,000 = $0.133 — exactly the per-image figure. That is the architectural difference from gpt-image-2, which uses a newer calculator-based token model rather than these fixed legacy counts. [src: https://developers.openai.com/api/docs/guides/image-generation]
One thing people assume wrong: gpt-image-1.5 is not categorically cheaper than gpt-image-2. It is cheaper only at square medium and high quality — $0.034 vs gpt-image-2’s $0.053 at medium, and $0.133 vs $0.211 at high (1024×1024). [src: https://developers.openai.com/api/docs/guides/image-generation] At low quality and at every non-square size, gpt-image-2 is the cheaper one. So do not choose 1.5 to save money; choose it for transparency. For the bigger picture on OpenAI’s full pricing including ChatGPT subscriptions and API token costs, see the OpenAI pricing plans breakdown. There is no free image tier — every generation is billed, and gpt-image-1.5’s rate-limit table does not even offer a Free tier.
Rate limits for gpt-image-1.5
OpenAI publishes gpt-image-1.5’s per-tier limits including IPM (images per minute), the axis specific to image models. Your ceiling rises as your account’s usage and spend qualify it for a higher tier:
| Tier | TPM | IPM (images/min) |
|---|---|---|
| Free | Not supported | Not supported |
| Tier 1 | 100,000 | 5 |
| Tier 2 | 250,000 | 20 |
| Tier 3 | 800,000 | 50 |
| Tier 4 | 3,000,000 | 150 |
| Tier 5 | 8,000,000 | 250 |
[src: https://developers.openai.com/api/docs/models/gpt-image-1.5]
The practical read: a fresh Tier 1 account is capped at 5 images per minute, so if you are batching cut-outs you will feel it early. For how to read the live rate-limit headers on each response and branch on error codes, see the OpenAI image generation API guide — that covers the full request/response contract this page does not duplicate.
Calling gpt-image-1.5: the minimal request
Calling gpt-image-1.5 is the same shape as any GPT Image call — set the model string and, for its signature use case, ask for a transparent background:
1 | from openai import OpenAI |
Two setup notes. First, GPT Image models — gpt-image-1.5 included — may require you to complete API Organization Verification in your developer console before they will run; build that into setup, not launch day. [src: https://developers.openai.com/api/docs/guides/image-generation] Second, this is deliberately the only code here. The full parameter tables, the edits endpoint with masks, the request/response JSON, and the error-handling patterns are in the OpenAI image generation API guide. No point repeating them per model.
For calling GPT Image 1.5 through a unified API rather than OpenAI directly, Modellix exposes it at POST /api/v1/openai/gpt-image-1.5/async — same model, same output, billed per image instead of per token. The async contract (submit → poll task_id → retrieve result) is identical across all models on the platform. [src: https://docs.modellix.ai/ways-to-use/api] The Modellix pricing for gpt-image-1.5 runs $0.0138–$0.2334 per image depending on resolution and quality, and the edit variant at the same range. [src: https://www.modellix.ai/models/openai/gpt-image-1.5]
When to actually choose gpt-image-1.5
Here is the rule, stated as narrowly as the facts allow. Use gpt-image-1.5 only if you need at least one of: a transparent background, strict fidelity control, or merging more than one reference image (up to 16) — and you can migrate off it before December 1, 2026. For everything else — new builds, higher resolution, longest support window, lower cost at non-square sizes — gpt-image-2 is the better and longer-lived choice.
Even inside its niche, treat gpt-image-1.5 as temporary. If transparent PNGs are core to your product, the honest plan is: ship on gpt-image-1.5 now if you must, but track OpenAI’s transparency support on newer models and be ready to move. The model you are depending on has a published end date.
Migrating later is mostly a one-line change — swap model="gpt-image-1.5" for model="gpt-image-2" — with two things to watch. First, any code path that sets background="transparent" will start failing on gpt-image-2, since that model rejects the parameter; you will need an alternative for alpha output before you flip the switch, whether that is post-processing to key out a solid background or keeping a separate transparency pipeline. Second, the cost profile shifts: gpt-image-2 meters output through a newer token calculator rather than gpt-image-1.5’s fixed legacy counts, and it is cheaper at low quality and at non-square sizes but pricier at square high quality, so re-check your per-image budget against the sizes you actually request rather than assuming a flat increase. Neither is hard, but both are the kind of thing you want to discover in a planned migration, not on December 1.
One key across models while you compare
If you are weighing gpt-image-1.5 against gpt-image-2 — or against Google’s or other providers’ image models — an overview of production text-to-image models helps narrow the field. An aggregation layer lets you call them all on one key while you decide, which saves re-integrating each SDK during evaluation. Modellix carries four OpenAI image models — gpt-image-1.5 and gpt-image-2 plus their edit variants — metered per image, with per-call cost and latency logged so you can compare like for like.
Two honest caveats. That $0.0138 low end sits above OpenAI’s own $0.009 low-quality rate, so this is not a cheaper source — it is a distribution layer, priced per call with the model owner’s economics passed through. And it carries those four GPT Image models, not gpt-image-1, gpt-image-1-mini, DALL·E, or Sora — for anything outside that set you would go to OpenAI directly. What a shared key buys you is one integration while you compare models, not a lower unit price.
FAQ
Is gpt-image-1.5 still available through the API?
Yes, for now. It is deprecated but still callable until its shutdown date of December 1, 2026, after which the model string will stop working. OpenAI recommends gpt-image-2 as the replacement. [src: https://developers.openai.com/api/docs/deprecations]
gpt-image-1.5 vs gpt-image-2 — which should I use?
gpt-image-2 for almost everything: it is newer, not deprecated, and supports up to 4K. Choose gpt-image-1.5 only when you specifically need a transparent background, strict fidelity control, or merging up to 16 reference images — and can migrate before December 1, 2026.
Does gpt-image-1.5 support transparent backgrounds?
Yes — natively. That is its main advantage over gpt-image-2, which does not support background: "transparent". Pass background="transparent" to get an alpha-channel PNG. [src: https://developers.openai.com/api/docs/guides/image-generation]
What does gpt-image-1.5 cost per image?
At 1024×1024: $0.009 (low), $0.034 (medium), $0.133 (high). Non-square sizes cost more — up to $0.20 for a high-quality 1536×1024. There is no free image tier. [src: https://developers.openai.com/api/docs/models/gpt-image-1.5]
What are gpt-image-1.5’s rate limits?
They scale by usage tier: Tier 1 is 5 images per minute, rising to 250 at Tier 5; the Free tier does not support the model. Token-per-minute limits run from 100,000 (Tier 1) to 8,000,000 (Tier 5). [src: https://developers.openai.com/api/docs/models/gpt-image-1.5]
When does gpt-image-1.5 shut down?
December 1, 2026. It was deprecated on June 2, 2026, with gpt-image-2 as the recommended replacement. Plan any migration before that date. [src: https://developers.openai.com/api/docs/deprecations]
