AI Image Generation API Pricing Comparison: GPT Image, Gemini Image, and Imagen Units

AI image generation API pricing is hard to compare because providers do not all sell the same unit. OpenAI GPT Image uses token-based image cost estimates, Google Gemini image models blend text, input image, and output image token pricing, and Google Imagen is often shown as a direct per-image price. If you compare only the headline number, you can miss input-image charges, output resolution tiers, batch/flex discounts, retries, and route-specific catalog status.

This guide normalizes the current units behind GPT Image, Gemini Image, and Imagen, then shows how to check the same rows in Flatkey pricing. The source check was run on June 17, 2026 Asia/Shanghai using official OpenAI image generation pricing guidance, Google Gemini API pricing, Google Vertex AI pricing, and a live Flatkey public pricing-page snapshot. Treat every price below as a point-in-time planning aid and verify the current provider and Flatkey dashboard rows before production traffic.

Quick Answer: How To Compare AI Image Generation API Pricing

Use four columns whenever you compare AI image generation API pricing:

1. Generation unit: token-based image output, per generated image, per input image, or a dashboard-specific routed unit.
2. Resolution and quality: square, landscape, portrait, 1K, 2K, 4K, low, medium, high, standard, fast, or ultra.
3. Extra inputs: prompt text, reference images, masks, edits, grounding/search, partial images, and retries.
4. Route evidence: exact Flatkey model row, endpoint type, status, usage logs, and current pricing label.

In practical terms: GPT Image needs token and quality math, Gemini Image needs token-to-image-equivalent math, and Imagen needs per-image plus deprecation and feature-route checks. Flatkey can help centralize model rows, endpoint support, and usage logs, but the selected row still needs verification.

Pricing Unit Comparison

The table below is the core normalization view for AI image generation API pricing. It intentionally avoids ranking the models by "cheapest" because each provider exposes different units and each model targets a different workflow.

Provider Family	Current Official Unit	Example Public Price Points Checked	Hidden Cost Checks	Flatkey Check
OpenAI GPT Image 2	Image output token estimate plus text and image input tokens.	OpenAI's image guide calculator lists GPT Image 2 at $0.006, $0.053, and $0.211 for 1024x1024 low, medium, and high examples; 1536x1024 medium is listed at $0.041.	Prompt tokens, reference-image input tokens, quality, size, output format, partial-image streaming, retries, and edits.	Check gpt-image-2 or openai/gpt-image-2, endpoint type, quota type, current pricing label, and status.
Google Gemini 3 Pro Image	Text/image input per 1M tokens and image output per 1M image tokens, expressed as image equivalents.	Google AI lists paid standard input at $2.00 per 1M text/image tokens and image output at $120 per 1M tokens, equivalent to $0.134 per 1K/2K image and $0.24 per 4K image.	Input image tokens, output resolution, web/image search grounding charges, tier type, and whether batch/flex or priority applies.	Check gemini-3-pro-image and gemini-3-pro-image-preview rows, endpoint type, and availability status.
Google Gemini 2.5 Flash Image	Text/image input per 1M tokens and output shown as a per-image equivalent.	Google AI lists standard input at $0.30 per 1M text/image tokens and output at $0.039 per 1024x1024 image; batch/flex output is $0.0195 per image.	Preview-model behavior, batch/flex eligibility, input tokens, output size, retries, and route support.	Flatkey's June 17 snapshot showed gemini-2.5-flash-image as available with gemini and openai endpoint types.
Google Imagen 4	Per generated image in the Gemini API and Vertex AI pricing tables.	Google AI lists Imagen 4 Fast at $0.02, Standard at $0.04, and Ultra at $0.06 per image.	Deprecation timeline, feature type, upscaling, editing, product recontext, virtual try-on, and route status.	Flatkey's June 17 snapshot listed Imagen 4 rows with image-generation, gemini, and openai endpoint types, but with unknown_failure status.

OpenAI GPT Image Pricing: Token Math First

OpenAI's current image generation guide says GPT Image models can generate and edit images through the Image API and can also be used through the Responses API image-generation tool. For gpt-image-2, the official cost section says to estimate output tokens from requested quality and size, then account for text input tokens and image input tokens if the request edits or references images.

That makes GPT Image different from a simple per-image SKU. The visible price for a generated asset can change when you change quality, size, input images, streaming partials, or retry behavior. OpenAI's calculator examples show GPT Image 2 at these point-in-time estimates:

GPT Image 2 Setting	1024x1024	1024x1536	1536x1024	Planning Note
Low	$0.006	$0.005	$0.005	Useful for drafts, thumbnails, and fast iteration.
Medium	$0.053	$0.041	$0.041	A practical default to benchmark before final quality.
High	$0.211	$0.165	$0.165	Use for final assets, not every preview.

Two details matter for AI image generation API pricing: OpenAI notes that edit requests with reference images can use more input tokens, and each partial image in streaming adds 100 image output tokens. For production budgets, separate "preview" and "final render" settings rather than letting every user action call a final-quality route.

Gemini Image Pricing: Token Prices With Image Equivalents

Google's Gemini API pricing page separates native image-generation models from general text models. Gemini image models still use token pricing under the hood, but the pricing page often gives an equivalent per-image number so teams can plan without manually multiplying image output tokens.

For Gemini 3 Pro Image, Google AI lists the standard paid input price at $2.00 per 1M text/image tokens and says image input is set at 560 tokens, or $0.0011 per image. The same page lists image output at $120 per 1M tokens, with examples of $0.134 per 1K/2K output image and $0.24 per 4K output image. Batch/flex examples cut those output examples to $0.067 and $0.12 respectively.

For Gemini 2.5 Flash Image, Google AI lists standard paid input at $0.30 per 1M text/image tokens and output at $0.039 per image. The footnote says image output is priced at $30 per 1M tokens, with 1024x1024 output images consuming 1290 tokens. The same page lists batch and flex output at $0.0195 per image and priority at $0.0702 per image.

On Google Cloud Vertex AI pricing, Gemini image pricing can appear with different model naming and route context. For example, the Vertex page lists Gemini 3 Pro Image image output at $120 per 1M tokens and includes a footnote mapping 1120 tokens to $0.134 for 1K/2K and 2000 tokens to $0.24 for 4K. It also lists Gemini 3.1 Flash Image output examples by resolution. If your team uses Vertex AI instead of the Gemini API, do not assume the same route, model name, or tier behavior as the Gemini API page.

Imagen Pricing: Per Image, But Watch Deprecation And Feature Type

Imagen looks simpler because Google lists direct per-image prices. On the Gemini API pricing page checked for this article, Imagen 4 Fast is $0.02 per image, Imagen 4 Standard is $0.04 per image, and Imagen 4 Ultra is $0.06 per image. The Vertex AI pricing page also lists Imagen 4 Fast at $0.02, Imagen 4 at $0.04, and Imagen 4 Ultra at $0.06 per generated image.

The caveat is freshness. Google's Gemini API pricing page includes a warning that Imagen 4 models imagen-4.0-generate-001, imagen-4.0-ultra-generate-001, and imagen-4.0-fast-generate-001 are deprecated and scheduled to shut down on August 17, 2026, with migration guidance toward Gemini 2.5 Flash Image. That warning is a bigger operational issue than a one-cent price difference.

Feature type also matters. Vertex AI lists separate Imagen prices for upscaling, visual captioning, visual Q&A, product recontext, and virtual try-on. A "per image" price for text-to-image generation is not automatically the same price as an edit, upscaling step, or product-image workflow.

Flatkey Catalog Snapshot For Image Models

Flatkey's public pricing page was checked on June 17, 2026 Asia/Shanghai. The server-rendered pricing page reported 638 AI models and pricing version 5a90f2b86c08bd983a9a2e6d66c255f4eaef9c4bc934386d2b6ae84ef0ff1f1f. The same page exposed endpoint families including gemini, image-generation, openai, and openai-response. That matters because AI image generation API pricing through a gateway must match both the model row and the endpoint type.

Flatkey Row Checked	Endpoint Types	Pricing Fields Seen	Status In Snapshot	Action Before Production
gpt-image-2	openai	model_ratio: 3.325, completion_ratio: 6, image_ratio: 1.6	unknown_failure	Confirm whether the selected row supports your intended image request shape before any launch.
openai/gpt-image-2	openai	quota_type: 1, model_price: 0.063	unknown_failure	Do not copy this number without confirming current dashboard unit and route behavior.
gemini-2.5-flash-image	gemini, openai	model_ratio: 0.15, completion_ratio: 100	available	Run a small route test and confirm whether the app should call the Gemini-style or OpenAI-style endpoint.
gemini-3-pro-image-preview	gemini, openai	model_ratio: 1, completion_ratio: 60	available	Confirm preview/stable naming and rate limits before building a long-lived route.
imagen-4.0-fast-generate-001, imagen-4.0-generate-001, imagen-4.0-ultra-generate-001	image-generation, gemini, openai	model_ratio: 37.5, completion_ratio: 1	unknown_failure	Treat as dated catalog presence only; also account for Google's August 17, 2026 Imagen 4 shutdown warning.

This Flatkey evidence does not mean every image row is production-ready today. It means the current catalog can be used as a starting point for provider comparison, route selection, quota review, and logging. The actual production gate is a current dashboard check plus a small smoke test.

A Normalization Formula For Image Cost Planning

Use this normalized formula before choosing a provider or route:

Estimated image workflow cost =
  prompt text cost
+ input image or edit cost
+ generated image output cost
+ partial image or preview cost
+ retry and failure cost
+ gateway or route-specific unit cost

For GPT Image, the generated image output cost is driven by image output tokens. For Gemini Image, the published image-equivalent numbers are derived from token prices and expected token counts. For Imagen, the generated image cost is usually visible as a per-image price, but edits and related image features may have separate prices. Flatkey then adds the operational layer: exact row, endpoint support, group, quota type, route status, and actual usage log.

Procurement Checklist For Image Generation APIs

A team comparing AI image generation API pricing should collect these fields before approving a model for production:

Field	Why Finance Cares	Why Engineering Cares
Exact model ID	Prevents budgeting against the wrong SKU.	Prevents calls to aliases, previews, or deprecated rows.
Endpoint path	Confirms which usage ledger will be billed.	Confirms request body, SDK behavior, and response shape.
Output resolution and quality	Separates preview spend from final-render spend.	Controls latency, UX, retry pressure, and storage.
Input-image handling	Edits and reference images can add cost.	Masks, references, and image uploads change request structure.
Deprecation or preview status	A cheap route can become a migration cost.	Preview and deprecated routes need fallbacks.
Flatkey usage logs	Shows actual cost after retries and failures.	Shows model row, route status, error rate, and rollback signal.

Template: Compare One Prompt Across Routes

Use one controlled prompt and one output setting when you benchmark routes. Change only one variable at a time.

Benchmark run
- Prompt category: product concept, marketing asset, UI mockup, or edit
- Output target: draft, preview, or final
- Size: fixed across providers where possible
- Quality/tier: fixed or documented as provider-specific
- Input images: none, one reference, or edit with mask
- Route A: direct provider baseline
- Route B: Flatkey selected row and endpoint
- Log: status, latency, retries, output accepted, usage unit, final cost

This gives you a real AI image generation API pricing comparison instead of a spreadsheet that ignores rejected images, moderation blocks, retries, and routing differences.

When Flatkey Helps With Image Pricing

Flatkey is useful when image generation is part of a larger multimodal workflow. A production app might use one model for prompt rewriting, GPT Image or Gemini Image for generation, an Imagen row for a specific creative route, and a vision-capable model for review. Flatkey's value is the shared API access layer: one key, model-row visibility, endpoint choices, usage logs, quotas, and a single pricing surface.

That does not remove provider-level due diligence. It makes the checks easier to repeat. For AI image generation API pricing, the safest Flatkey workflow is:

1. Open Flatkey pricing and search the exact model row.
2. Confirm provider family, endpoint types, group, quota type, and current price label.
3. Compare against official OpenAI or Google pricing docs for the same model and route.
4. Run one low-risk smoke test through the confirmed endpoint.
5. Review Flatkey logs for model row, status, usage unit, and cost.
6. Set quota limits before exposing the feature to production users.
7. Document fallback routes for preview models or deprecated models.

Related Pricing Guides

If you are building a broader cost model, pair this page with OpenAI Image API Pricing, AI Model Pricing Comparison, and GPT Image 2 API Through Flatkey. Those articles cover provider-specific checks and broader model pricing patterns; this guide focuses on cross-provider image pricing units.

FAQ

What is the best way to compare AI image generation API pricing?

Compare normalized units, not just headline prices. Record prompt text cost, input image cost, generated output cost, resolution, quality, retry behavior, and the exact Flatkey or provider route used.

Is GPT Image priced per image?

OpenAI's current GPT Image 2 guidance uses output-token estimates from quality and size, plus text and image input tokens. The guide provides per-image calculator examples, but the cost model is token-driven.

Is Gemini Image priced per image?

Google lists token prices and image-equivalent outputs. Gemini 3 Pro Image uses image output token pricing with examples such as $0.134 for 1K/2K standard outputs, while Gemini 2.5 Flash Image lists output examples such as $0.039 per 1024x1024 image.

Is Imagen priced per image?

Google's Gemini API and Vertex AI pricing pages list Imagen 4 generation as direct per-image prices, such as $0.02 for Fast, $0.04 for Standard, and $0.06 for Ultra. Check feature type and deprecation status before relying on Imagen 4 for new production work.

How should I check AI image generation API pricing in Flatkey?

Search the exact model row in Flatkey pricing, confirm endpoint type, group, quota type, availability status, and current pricing label, then run a small smoke test and compare the usage log against your estimate.

Final recommendation: treat AI image generation API pricing as a workflow cost, not a single model price. Start with official provider units, normalize the request settings, then use Flatkey to verify the model row, endpoint, usage logs, and quota controls. View Pricing before moving image traffic into production.