GuidesOneMux

GPT Image 2 API costs for structured extraction: an assumption-driven estimate

Estimate GPT Image 2 workload costs using explicit token assumptions, and understand why JSON extraction support must be verified before budgeting a production workflow.

AI

Summary · TL;DR

At the third-party rates of $5 per million input tokens and $30 per million output tokens, an illustrative request using 3,000 input and 400 output tokens costs $0.027. Structured JSON output is not advertised in the supplied model source, so feasibility and actual billing must be verified before treating this as a production estimate.

Key Takeaways

  • The illustrative workload costs $27 per 1,000 requests, or $4,050 for 150,000 requests in a 30-day month, before retries and other charges.
  • The OneMux catalogue snapshot lists GPT Image 2 prices and its discount ratio as unknown; a third-party price is not a OneMux quote.
  • The supplied third-party source describes GPT Image 2 as an image-generation model and does not advertise structured output.
  • Validate extraction output locally against a JSON schema and budget for failed attempts, but first confirm that the chosen endpoint can return the required text.
On this page (8)

The decision in brief

For an illustrative token-billed workload, GPT Image 2 would cost $0.027 per request, or $27 per 1,000 requests, at 3,000 input tokens and 400 output tokens per request. That uses third-party rates of $5 per million input tokens and $30 per million output tokens—not a OneMux price quote.

There is a more fundamental qualification: the supplied source describes GPT Image 2 as an image-generation model and does not advertise structured output. Vision input alone does not establish that an endpoint can extract fields into JSON. Before using this estimate for a structured-extraction project, confirm that the intended endpoint accepts the document and returns text suitable for validation. Otherwise, select a model and endpoint documented for that workflow.

This is a cost guide from OneMux, the publisher. OneMux provides access to models through one OpenAI-compatible API, but compatibility with a particular endpoint and its parameters must be confirmed in the applicable documentation.

Price evidence and its limits

The OneMux published catalogue snapshot was captured on November 18, 2026. Its GPT Image 2 price fields were unknown; unknown does not mean free. The OneMux model page is useful for checking the current listing.

Model and price recordOfficial inputOfficial outputOneMux inputOneMux outputPublished OneMux discount ratio
GPT Image 2, OneMux catalogue snapshotUnknownUnknownUnknownUnknownUnknown
GPT Image 2 (2026-04-21), third-party listing$5.00 per 1 million tokens$30.00 per 1 million tokensNot statedNot statedNot stated

The second row is a separate record, not a replacement for the missing OneMux prices. Future AGI's listing attributes its rates to LiteLLM and reports that they were verified on October 22, 2026. It also lists cached input at $1.25 per million tokens. Confirm current prices, model identifier, modalities and billing rules with the provider and route you intend to use. The source does not establish that those rates or caching terms apply on OneMux.

The source also leaves context and maximum input/output limits unspecified, while the OneMux catalogue lists a 1M context. Treat that discrepancy as unresolved rather than assuming either value applies to a particular endpoint.

Reproduce the workload estimate

Assume a 30-day month and the following hypothetical usage. The token counts include whatever input and output the billed endpoint reports; they are not a prediction of how an image or PDF will be tokenized.

AssumptionValue
Requests per day5,000
Requests in 30 days150,000
Billed input per request3,000 tokens
Billed output per request400 tokens
Input rate$5.00 per 1 million tokens
Output rate$30.00 per 1 million tokens
Cached input, retries and other chargesExcluded

For each request:

  • Input: 3,000 ÷ 1,000,000 × $5 = $0.015.
  • Output: 400 ÷ 1,000,000 × $30 = $0.012.
  • Total: $0.015 + $0.012 = $0.027.

That produces the following comparable units:

VolumeEstimated cost
1 request$0.027
1,000 requests$27
5,000 requests, or one assumed day$135
150,000 requests, or 30 assumed days$4,050

The source's calculator shows approximately $4,109 for about 152,188 monthly requests. That is compatible with the same $0.027 per-request assumption: 152,188 × $0.027 is approximately $4,109.08. The difference from $4,050 is the assumed request count, not a different unit price.

Allow for retries and cache separately

If 5% of the 150,000 initial requests each require one additional attempt of the same billed size, that adds 7,500 attempts × $0.027 = $202.50. The estimated monthly total becomes $4,252.50. This is an assumption, not an observed failure rate; longer repair prompts or responses would change it. Local schema validation itself is not a model request, but a model retry is.

Caching can change the input portion only when the route actually supports it and reports eligible cached tokens. As an illustration—not a OneMux quote—suppose 1,000 of the 3,000 input tokens per request are billed at the third-party listing's $1.25-per-million cached rate, while the remaining 2,000 are billed at $5. Input then costs $0.01125; adding the unchanged $0.012 output gives $0.02325 per request, or $3,487.50 for 150,000 requests. The source lists a cached rate but does not advertise prompt caching, so verify eligibility and billing before using this scenario in a budget.

Plan the extraction and validation workflow

The following is an illustrative workflow, not a tested GPT Image 2 integration. Its prerequisite is an endpoint documented to accept your image or PDF and return text. If that is unavailable, use an appropriately documented extraction model instead; image generation is not a substitute for field extraction.

  1. Define the output contract. For an invoice, a minimal JSON Schema might require an object with a string invoice_number, a string currency, and a numeric total, with all three fields required. Decide whether missing or unreadable values should be rejected or represented explicitly.
  2. Provide the document and instructions. Ask the supported endpoint to return only the specified fields. Use a native schema-constrained output parameter only if that endpoint documents it; the supplied GPT Image 2 source does not advertise structured output.
  3. Parse and validate locally. Check that the response is valid JSON, then validate it against your schema with a JSON Schema validator. A plausible-looking response may still have missing keys or incorrect types.
  4. Apply business checks. Schema validity does not prove that an extracted total matches the document. Flag ambiguous values and reconcile important fields against the source or a review process.
  5. Record usage and outcomes. Track billed input and output, validation failures, retries and successful documents. Divide total spend by successfully accepted documents as well as by requests.

Common failures include non-JSON responses, required fields omitted, amounts returned as strings, and valid JSON containing incorrect values. Set a retry limit and send unresolved cases for review rather than assuming another call will fix them. Include review costs and any separate OCR or downstream processing in the full workflow budget; they are outside the token estimate above.

Before committing a budget

Run a representative sample only after confirming endpoint support and current rates. Use the resulting billed usage—not character counts or an assumed image-to-token conversion—to replace the example's 3,000 and 400 tokens. Check whether failed calls are billed, whether cached tokens qualify, and whether the route has tiers or modality-specific charges. The supplied GPT Image 2 records do not establish OneMux tier or cache prices.

The practical conclusion is conditional: $27 per 1,000 requests is reproducible under the stated assumptions, but it does not establish either OneMux pricing or GPT Image 2's suitability for JSON extraction. Validate the workflow first, then calculate its cost per accepted document.

Sources

Published price snapshot

Captured 2026-11-18T15:30:05.993Z from the OneMux published catalogue. Prices retain website billing units; this is not a live benchmark. Tiered, cached or time-based billing may differ. Unknown values are not zero.

ModelOfficial inputOfficial outputOneMux inputOneMux outputPublished price ratio
GPT Image 2UnknownUnknownUnknownUnknownUnknown / not discounted

Current OneMux prices. Compare workload costs only when input/output and billing units match; see each model page for tier conditions.

FAQ

How much does GPT Image 2 cost per 1,000 requests?

Under the illustrative third-party rates and an assumption of 3,000 input tokens and 400 output tokens per request, the cost is $27 per 1,000 requests before retries or other charges. This is not a OneMux price quote.

Does GPT Image 2 support JSON schema-constrained output?

The supplied third-party listing does not advertise structured output. Confirm that your specific endpoint returns text and supports any required schema parameter before designing the workflow around it.

What is the OneMux discount for GPT Image 2?

The November 18, 2026 OneMux catalogue snapshot lists the model's prices and discount ratio as unknown. Unknown values should not be treated as zero or as evidence of a discount.

Related articles