# GPT Image 1.5 Pricing in 2026: Cost per Image and the Move to 2.5

> GPT Image 1.5 costs $0.009, $0.034, and $0.133 per 1024x1024 image and shuts down December 1, 2026. Per-image prices for every GPT Image model, plus 2.5 rates.

- Source: https://www.aifreeapi.com/en/posts/gpt-image-1-5-pricing
- Language: en
- Published: 2026-03-23
- Updated: 2026-10-09
- Publisher: AI Free API (https://www.aifreeapi.com)

**As of October 9, 2026, OpenAI's [gpt-image-1.5 model page](https://developers.openai.com/api/docs/models/gpt-image-1.5) lists GPT Image 1.5 at $0.009, $0.034, and $0.133 per 1024x1024 image at low, medium, and high quality.** Those prices still apply, but the model is now deprecated: OpenAI's [deprecations page](https://developers.openai.com/api/docs/deprecations) shuts down `gpt-image-1.5`, `gpt-image-1-mini`, and `chatgpt-image-latest` on **December 1, 2026**, and `gpt-image-1` on **October 23, 2026**. The recommended replacement for all four is `gpt-image-2.5-sunburst` or `gpt-image-2.5-flare`.

Three facts shape any budget built on these models today:

- The cheapest documented per-image price in the GPT Image family is **$0.005**, shared by `gpt-image-1-mini` (low, 1024x1024) and `gpt-image-2` (low, 1024x1536 or 1536x1024). Mini disappears on December 1, 2026; `gpt-image-2` is not deprecated, so after that date it holds the floor alone.
- GPT Image 2.5 has **no official per-image price table**. OpenAI publishes token rates for it ($8 per million image input tokens, $2 cached, and **$30 per million image output tokens**) plus an interactive calculator in the image generation guide. Any fixed per-image dollar figure you see for 2.5 is an estimate.
- GPT Image 1.5's $32 output rate drops to $30 on 2.5, about a 6% cut on the rate, but the bill can still go up. GPT Image 2 bills the same $30 rate, and its official price for a medium 1024x1024 image is $0.053, against $0.034 on 1.5, because it spends more tokens per image.

If you only need the 1.5 number, the next section has the full table. If you are deciding what to bill on through the end of 2026, skip to the model-by-model table and the migration section.

![GPT Image 1.5 pricing overview](https://www.aifreeapi.com/posts/en/gpt-image-1-5-pricing/img/cover.webp)

## GPT Image 1.5 pricing per image: $0.009, $0.034, and $0.133 at 1024x1024

The official per-image table on the gpt-image-1.5 model page has six cells, three quality levels across two shapes:

| Quality | 1024x1024 | 1024x1536 or 1536x1024 |
|---|---:|---:|
| Low | $0.009 | $0.013 |
| Medium | $0.034 | $0.05 |
| High | $0.133 | $0.20 |

`chatgpt-image-latest`, the alias for the image snapshot ChatGPT previously used, has an identical table on its own [model page](https://developers.openai.com/api/docs/models/chatgpt-image-latest), so switching between the two IDs never changes the price. It shares the December 1, 2026 shutdown date as well.

Underneath the table, GPT Image 1.5 bills by tokens. The [OpenAI pricing page](https://developers.openai.com/api/docs/pricing) lists image tokens at **$8 per million input, $2 cached input, and $32 output**, and text tokens at $5 input, $1.25 cached, and $10 output. The per-image prices are that token math done for the common sizes. OpenAI's [image generation guide](https://developers.openai.com/api/docs/guides/image-generation) gives the output token counts for a 1024x1024 image on these earlier models: **272** tokens at low, **1,056** at medium, and **4,160** at high. Multiply by $32 per million and you land on the table:

```text
1,056 output tokens x $32 / 1,000,000 = $0.0338  -> listed as $0.034 (medium)
4,160 output tokens x $32 / 1,000,000 = $0.1331  -> listed as $0.133 (high)
```

That check matters later, because it is the only way to estimate GPT Image 2.5, which has token rates but no table.

The release date behind this model: the only snapshot is `gpt-image-1.5-2025-12-16`, so GPT Image 1.5 dates from **December 16, 2025**. It stays callable until December 1, 2026.

## Every GPT Image model's price per image: 1, 1-mini, 1.5, 2, and 2.5

Five GPT Image models have official per-image tables; the two GPT Image 2.5 models have only token rates. Reading them side by side is the fastest way to answer "what does GPT Image cost" for whichever model ID you actually call. All figures are OpenAI list prices as of October 9, 2026, Standard (non-Batch) processing.

| Model ID | Low 1024x1024 | Medium | High | Image output tokens | Status |
|---|---:|---:|---:|---:|---|
| `gpt-image-1-mini` | **$0.005** | $0.011 | $0.036 | $8 / M | Shuts down December 1, 2026 |
| `gpt-image-1.5` | $0.009 | $0.034 | $0.133 | $32 / M | Shuts down December 1, 2026 |
| `chatgpt-image-latest` | $0.009 | $0.034 | $0.133 | $32 / M | Shuts down December 1, 2026 |
| `gpt-image-1` | $0.011 | $0.042 | $0.167 | $40 / M | Shuts down October 23, 2026 |
| `gpt-image-2` | $0.006 | $0.053 | $0.211 | $30 / M | Current (released April 21, 2026) |
| `gpt-image-2.5-flare` | no official table | no official table | no official table | $30 / M | Current (released September 8, 2026) |
| `gpt-image-2.5-sunburst` | no official table | no official table | no official table | $30 / M | Current (released September 8, 2026) |

Sources for each row: the GPT Image 2 per-image prices come from the ["Calculating costs" section of the image generation guide](https://developers.openai.com/api/docs/guides/image-generation#calculating-costs); the others come from the model pages for [gpt-image-1-mini](https://developers.openai.com/api/docs/models/gpt-image-1-mini), [gpt-image-1](https://developers.openai.com/api/docs/models/gpt-image-1), [gpt-image-2](https://developers.openai.com/api/docs/models/gpt-image-2), [gpt-image-2.5-flare](https://developers.openai.com/api/docs/models/gpt-image-2.5-flare), and [gpt-image-2.5-sunburst](https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst), plus the pricing page for token rates and the [API changelog](https://developers.openai.com/api/docs/changelog) for release dates (GPT Image 2 on April 21, 2026; the two 2.5 models on September 8, 2026).

Portrait and landscape (1024x1536 or 1536x1024) prices for the models with tables, low / medium / high: gpt-image-1-mini $0.006 / $0.015 / $0.052; gpt-image-2 $0.005 / $0.041 / $0.165; gpt-image-1 $0.016 / $0.063 / $0.25; gpt-image-1.5 and chatgpt-image-latest $0.013 / $0.05 / $0.20.

A few things the table makes obvious that the pricing page alone does not:

- **gpt-image-1 pricing** is the highest in the family at every quality level, and it is the first to go. If anything still calls `gpt-image-1`, it fails after October 23, 2026, two weeks from the date above.
- **GPT Image 2 is not cheaper than 1.5 everywhere.** Its output rate is lower ($30 versus $32 per million), and it wins at low quality and at every portrait or landscape size. At 1024x1024 medium and high it costs more: $0.053 versus $0.034, and $0.211 versus $0.133.
- The 2 and 2.5 models share one set of token rates. The [gpt-image-2.5-sunburst page](https://developers.openai.com/api/docs/models/gpt-image-2.5-sunburst) says "Token rates match GPT Image 2." The price difference between GPT Image 2, Flare, and Sunburst comes entirely from how many tokens each spends per image, and OpenAI publishes a per-image table only for GPT Image 2.
- DALL·E 2 and DALL·E 3 are gone; their snapshots were removed on May 12, 2026, so they no longer belong in a GPT Image price comparison.

For the question of which model to pick on features rather than price, the guide to [OpenAI image generation API models](/en/posts/openai-image-generation-api-models) covers size limits, quality options, and endpoints.

## Cheapest GPT Image model per image: $0.005 on 1-mini and gpt-image-2

![Compare actual request charges and retries against the number of usable images](https://www.aifreeapi.com/posts/en/gpt-image-1-5-pricing/img/cost-per-usable-image.webp)

The cheapest documented price for any GPT Image model is **$0.005**, and two models share it: `gpt-image-1-mini` at low quality and 1024x1024, and `gpt-image-2` at low quality and 1024x1536 or 1536x1024. Their full official tables:

| Quality | 1-mini 1024x1024 | 1-mini portrait or landscape | gpt-image-2 1024x1024 | gpt-image-2 portrait or landscape |
|---|---:|---:|---:|---:|
| Low | **$0.005** | $0.006 | $0.006 | **$0.005** |
| Medium | $0.011 | $0.015 | $0.053 | $0.041 |
| High | $0.036 | $0.052 | $0.211 | $0.165 |

Token rates for 1-mini are also the lowest: $2.50 per million image input, $0.25 cached, $8 output, with text at $2 input and $0.20 cached. Batch halves those to $1.25 / $0.13 / $4 for image tokens.

Two conditions limit that answer:

1. **Mini's half of the answer ends on December 1, 2026.** After that date the cheapest documented per-image price belongs to `gpt-image-2` alone: $0.005 at low quality in portrait or landscape, $0.006 square. Above low quality, gpt-image-2 costs several times what mini did ($0.053 versus $0.011 at medium 1024x1024). GPT Image 2.5 may land above or below gpt-image-2; with no table, you answer that by measuring your own token usage (method below).
2. **Cheapest per image is not cheapest per usable image.** OpenAI describes 1-mini as "a cost-efficient version of GPT Image 1." Edits on 1-mini only accept `input_fidelity: low`, and if a prompt needs two or three attempts on mini but one on 1.5, the $0.009 model wins. For a bulk job of low-stakes variants where any output passes, mini is the cheapest option at medium and high quality for the next seven weeks.

If you are standing up new code on mini this late, read [gpt-image-1-mini API in 2026: Start with the Images API, Not Responses](/en/posts/gpt-image-1-mini-api) first; it only makes sense for work that finishes before the shutdown.

### What fal and other resellers charge for GPT Image models per image

A reseller's per-image price is that reseller's price, not OpenAI's, and today the figures are thin:

- **fal.ai** lists "GPT Image 2 API (openai/gpt-image-2)" and its edit endpoint at "$1 per unit" on its [pricing page](https://fal.ai/pricing), with a note that resolution and quality "may affect the final cost" and no definition of a unit. It publishes no price for `gpt-image-1-mini`, `gpt-image-1.5`, or `gpt-image-2.5`. So there is no fal per-image dollar figure for GPT Image 2 to set against OpenAI's official $0.006 / $0.053 / $0.211.
- Third-party price trackers quote per-image GPT Image 1.5 figures for other hosts (one lists $0.0030 per image on Replicate and $0.034 on Together for 1024x1024). Those are the hosts' own prices or the tracker's conversions; only the $0.034 matches OpenAI's medium-quality list price.
- Per-image GPT Image 2 figures on aggregator sites ($0.006 / $0.053 / $0.211 for low / medium / high at 1024x1024) match OpenAI's official table. Any fixed per-image figure for GPT Image 2.5 is an estimate, since OpenAI publishes none.

If the task is to find the cheapest GPT Image route rather than the cheapest OpenAI list price, compare a reseller's stated per-image price against the formula in the next section using your own measured token counts, not against someone else's estimate.

## What actually gets billed beyond the output image

![Flow map showing GPT Image 1.5 invoice components from prompt text, input images, and output image tokens, with a warning that high-fidelity edits cost more](https://www.aifreeapi.com/posts/en/gpt-image-1-5-pricing/img/full-invoice-map.webp)

For a short text prompt that produces one new image, the per-image table is close to the real charge, because output tokens dominate. The table stops being enough as soon as a request carries images in, streams partial results, or goes through the Responses API.

```text
request cost
= prompt text tokens x text input rate
+ input image tokens x image input rate      (edits, reference images)
+ output image tokens x image output rate    (the per-image table)
+ 100 output tokens per streamed partial image
```

Each line comes from the [image generation guide](https://developers.openai.com/api/docs/guides/image-generation) and the pricing page:

- **Input images are billed as image input tokens.** On GPT Image 1.5 that is $8 per million. `input_fidelity: high` on the edits endpoint preserves input detail, which generally means more input tokens. For `gpt-image-2`, OpenAI says to omit `input_fidelity` entirely; the model always runs at high fidelity, so reference-image edits may use more input tokens than they did on 1.5. OpenAI's reference does not state how `gpt-image-2.5-flare` or `gpt-image-2.5-sunburst` treat `input_fidelity`, so run one edit and read `usage.input_tokens` rather than assume.
- **Streaming costs tokens.** Each partial image requested through `partial_images` adds 100 output tokens.
- **Cached input rates for GPT Image 2 and 2.5 apply only through the Responses API**, not through direct calls to `/v1/images/generations` or `/v1/images/edits`. A Responses request also bills the mainline text model's tokens on top of the image model's.

Every Images API response includes a `usage` object with `input_tokens`, `output_tokens`, and `input_tokens_details` split into image and text tokens. That object, multiplied by the rates above, is the exact charge for the request. It is the tool you need for the next section.

## GPT Image 2.5 pricing: $30 per 1M output tokens, no per-image table

The question every GPT Image 1.5 user has to answer before December 1, 2026 is what the same workload costs on `gpt-image-2.5-flare` or `gpt-image-2.5-sunburst`. The honest answer has a known part and an unknown part.

**Known: the rates.** From the pricing page and the two 2.5 model pages:

| Rate (per 1M tokens) | gpt-image-1.5 | gpt-image-2.5 (both) and gpt-image-2 |
|---|---:|---:|
| Image input | $8 | $8 |
| Cached image input | $2 | $2 (Responses API only) |
| Image output | **$32** | **$30** |
| Batch image input / cached / output | $4 / $1 / $16 | $4 / $1 / $15 |
| Text input / cached | $5 / $1.25 | $5 / $1.25 |

The output rate falls by $2 per million tokens, about 6%. That alone does not make 2.5 cheaper per image.

**Unknown: tokens per image.** The image generation guide warns that "equal token rates don't mean equal cost per image," and GPT Image 2 shows it: at the $30 rate that 2.5 shares, OpenAI's [official GPT Image 2 table](https://developers.openai.com/api/docs/guides/image-generation#calculating-costs) puts a medium 1024x1024 image at $0.053, versus $0.034 on 1.5. For 2.5 there is no static table, only the guide's interactive calculator ("Select a model, quality, and size to estimate output tokens and image output cost"), and the guide notes the models "can use different token counts for the same quality setting." The 272 / 1,056 / 4,160 counts above are documented for earlier models, not for 2.5. On top of that, 2.5 adds two quality levels that did not exist on 1.5, `xhigh` and `max`, and the gpt-image-2 family accepts free `WIDTHxHEIGHT` sizes (gpt-image-2 documents a 3840 px maximum edge), so there are more ways to spend tokens than the old six-cell table could hold.

Two reference points bracket the budget, with the assumptions spelled out. *If* a medium 1024x1024 image on 2.5 used the 1,056 output tokens documented for earlier models, it would cost 1,056 × $30 / 1,000,000 = **$0.0317**, versus $0.034 on 1.5; at high quality, 4,160 × $30 / 1,000,000 = **$0.1248** versus $0.133. *If* it spent tokens the way GPT Image 2 does, it would land near GPT Image 2's official $0.053 medium and $0.211 high. Both are estimates, not OpenAI prices for 2.5. The spread between them, roughly $0.03 to $0.05 for a medium square image, is the reason to measure before December 1.

### How to price your own migration before December 1

The reliable way to find your cost on 2.5 is to measure it, and the response already tells you everything you need:

1. Pick 20 to 50 prompts that represent the real workload, with the sizes and quality levels you use today.
2. Run each once on `gpt-image-1.5` and once on `gpt-image-2.5-flare` (and on `gpt-image-2.5-sunburst` if editing precision is why you were on 1.5). The guide positions Flare for fast everyday generation and Sunburst for workflows "where editing precision matters most."
3. Sum `usage.output_tokens` and `usage.input_tokens` per model from each response.
4. Cost per image on 2.5 = (output tokens × $30 + image input tokens × $8 + text tokens × $5) / 1,000,000. For Batch, use $15, $4, and the Batch text rates instead.
5. Compare the two totals, then compare the acceptance rate of the outputs. A model that costs 5% more per image but needs fewer retries is cheaper per usable image.

If the test shows the per-image cost moving in the wrong direction, the levers are quality level (drop a step before you drop a model), Batch for anything asynchronous, and avoiding streamed partials you do not display.

For API examples and the Flare-versus-Sunburst decision itself, see [GPT Image 2.5: Flare vs. Sunburst, Pricing, and API Examples](/en/posts/gpt-image-2-5). If you are weighing `gpt-image-2` instead, [GPT Image 2 API Pricing: Cost per Image and Project Budget](/en/posts/openai-image-generation-api-pricing) works through project budgets; its official per-image table above is the baseline, and the same $30 output rate applies.

### Shutdown dates for every deprecated GPT Image model

| Model ID | Deprecation announced | Shutdown | Replacement |
|---|---|---|---|
| `gpt-image-1` | April 22, 2026 | **October 23, 2026** | gpt-image-2.5-sunburst or gpt-image-2.5-flare |
| `gpt-image-1.5` | June 2, 2026 | **December 1, 2026** | gpt-image-2.5-sunburst or gpt-image-2.5-flare |
| `gpt-image-1-mini` | June 2, 2026 | **December 1, 2026** | gpt-image-2.5-sunburst or gpt-image-2.5-flare |
| `chatgpt-image-latest` | June 2, 2026 | **December 1, 2026** | gpt-image-2.5-sunburst or gpt-image-2.5-flare |
| `dall-e-2`, `dall-e-3` | November 14, 2025 | May 12, 2026 (done) | gpt-image-2 at the time |

`gpt-image-2` is not on the deprecations page. Code that hard-codes `chatgpt-image-latest` as a "follows ChatGPT" alias breaks on the same day as 1.5; the alias no longer tracks anything newer. The older comparison of [chatgpt-image-latest vs gpt-image-1.5](/en/posts/chatgpt-image-latest-vs-gpt-image-1-5) is only relevant now for understanding what each ID was pointing at before you replace both.

## Batch pricing, rate limits, and monthly budget math

Batch is the one official lever that halves the price without changing the model. On the pricing page, GPT Image 1.5 Batch rates for image tokens are **$4 input, $1 cached input, and $16 output** per million, exactly half of Standard. The 2.5 and 2 Batch rates are $4 / $1 / $15; gpt-image-1 is $5 / $1.25 / $20; gpt-image-1-mini is $1.25 / $0.13 / $4.

Monthly math for GPT Image 1.5 at 1024x1024, output tokens only, using the documented token counts:

| Volume | Medium, Standard | Medium, Batch | High, Standard | High, Batch |
|---|---:|---:|---:|---:|
| 1,000 images | about $34 | about $17 | about $133 | about $67 |
| 10,000 images | about $340 | about $170 | about $1,330 | about $665 |

Batch figures are 1,056 × $16 / 1,000,000 ≈ $0.0169 per medium image and 4,160 × $16 / 1,000,000 ≈ $0.0666 per high image. Add prompt text and any input images on top; those are small for generation and material for edits.

Throughput is the other half of the budget. OpenAI now groups accounts into usage tiers named **Free, Build, Launch, and Grow** in the [rate limits guide](https://developers.openai.com/api/docs/guides/rate-limits): Free is a $100-per-month usage cap in allowed geographies, Build unlocks after $5 of credit purchases ($500 per month), Launch after $100 ($5,000 per month), and Grow after $500 ($200,000 per month). The 2.5 and gpt-image-2 model pages list the same default image limits per tier:

| Usage tier | Tokens per minute | Images per minute |
|---|---:|---:|
| Build | 250,000 | 20 |
| Launch | 3,000,000 | 150 |
| Grow | 8,000,000 | 250 |

Your organization's actual values are under Settings → Organization → Limits, and they can differ from the defaults. At 20 images per minute on Build, a 10,000-image job takes over eight hours of continuous calls, which is a reason to use Batch independent of price.

On "free GPT Image 1.5": the Images API has no free image allowance. The Free usage tier is a monthly spend cap on paid usage in supported countries, not a quota of free images, and nothing on the gpt-image-1.5 page offers one.

## Official prices versus calculator and reseller numbers

Numbers for GPT Image 1.5 disagree across the web for a structural reason, not because the price moves. OpenAI's pricing page shows token rates only; the per-image tables live on the individual model pages (for GPT Image 2, in the guide's "Calculating costs" section); the token counts per image are in the image generation guide. Three pages, one price.

Calculators that quote "$5 in / $10 out" are showing GPT Image 1.5's **text** token rates, not its image rates ($8 / $32). Resellers that quote a flat per-image price are quoting their own route, which may be cheaper or more expensive than OpenAI's list and tells you nothing about what OpenAI bills. And ChatGPT's own image generation is priced by subscription plan, not per image; if your question is how much ChatGPT charges for images, the per-day picture is in [ChatGPT Go Image Generation Limit per Day: No Official Number](/en/posts/chatgpt-go-image-generation-limit) and the free-plan equivalent in [ChatGPT Free Plan Image Generation Limits 2026](/en/posts/chatgpt-image-generation-limit-free-plan).

The rule of thumb: for what OpenAI charges, trust the model page, the pricing page, and the `usage` object in your own responses, in that order of convenience and the reverse order of precision.

## FAQ

**What is the current official GPT Image 1.5 price?**

As of October 9, 2026, $0.009, $0.034, and $0.133 per 1024x1024 image at low, medium, and high quality, and $0.013, $0.05, and $0.20 for 1024x1536 or 1536x1024, per the gpt-image-1.5 model page. The model shuts down on December 1, 2026.

**When was GPT Image 1.5 released, and how long is it available?**

Its only snapshot is `gpt-image-1.5-2025-12-16`, so December 16, 2025. It was marked deprecated on June 2, 2026 and is removed on December 1, 2026.

**Is `chatgpt-image-latest` cheaper than GPT Image 1.5?**

No. Its model page lists the same per-image table and the same token rates, and it shuts down on the same day. Neither ID should be in new code.

**What is the cheapest GPT Image model per image?**

Two models share the lowest documented price, $0.005: `gpt-image-1-mini` at low quality and 1024x1024, and `gpt-image-2` at low quality and 1024x1536 or 1536x1024. Mini shuts down on December 1, 2026, which leaves `gpt-image-2` low ($0.005 portrait or landscape, $0.006 square) as the cheapest documented per-image price. GPT Image 2.5 has no per-image table, so its cost per image is whatever `usage.output_tokens` says multiplied by $30 per million.

**How much does GPT Image 2.5 cost per image?**

OpenAI publishes no per-image table for GPT Image 2.5, only token rates and an interactive calculator in the image generation guide. The rates are $8 per million image input tokens, $2 cached (Responses API only), and $30 per million output tokens, with Batch at $4 / $1 / $15. Estimate per image as output tokens × $30 / 1,000,000 and measure the token count on your own prompts. GPT Image 2, at the same rates, officially costs $0.006 / $0.053 / $0.211 per 1024x1024 image; that is a comparison point, not a 2.5 price.

**Does GPT Image 1.5 pricing include edits and reference images?**

No. The per-image table covers output tokens. Input images are billed at $8 per million image input tokens and prompt text at $5 per million, and each streamed partial image adds 100 output tokens.

**Does Batch really matter for GPT Image 1.5?**

Yes. Batch halves the image token rates to $4 / $1 / $16 per million, so a medium 1024x1024 image works out to about $0.017 instead of $0.034. The same half-price structure applies on 2.5 ($15 output), so a Batch pipeline carries over.

**Is there a free tier for GPT Image 1.5 API usage?**

There are no free GPT Image 1.5 images. OpenAI's Free usage tier is a $100-per-month spend cap on paid API usage in supported countries, and the gpt-image-1.5 model page lists no free allowance.