GuidesOneMux

GPT Image 2.6 Sunburst API Costs for Document Question Answering

Build a reproducible document question-answering budget using explicit image-input assumptions, while separating unknown answer-generation and attribution costs.

AI

Summary · TL;DR

A source-based cost worksheet for document question answering that separates a hypothetical image-input charge from unknown text-answer, retrieval, and attribution costs.

Key Takeaways

  • The supplied pricing source reports $5 per million text input tokens and $8 per million image input tokens for Sunburst; these are not confirmed OneMux rates.
  • At an illustrative 120 text input tokens and 1,024 image input tokens per question, the reported input rates yield $0.008792 per request before answer generation and other charges.
  • The supplied evidence does not establish that Sunburst supports document question answering with source attribution or provide a text-output rate for that workflow.
  • OneMux's published snapshot lists Sunburst input and output prices as unknown, so it cannot support an all-in OneMux cost estimate.
On this page (8)

The decision: budget the documented inputs, not an imagined complete answer

For document question answering with citations, do not treat GPT Image 2.6 Sunburst's image-output price as the cost of a text answer. The supplied evidence reports image-model token rates but does not establish that Sunburst can answer document questions with source attribution, which endpoint supports that workflow, or what text-output rate would apply. Verify those points before choosing it for production.

You can still make a useful partial estimate. At the rates reported by Ofox's Image 2.6 pricing guide, a hypothetical request containing 120 text input tokens and 1,024 image input tokens costs $0.008792 in input charges. That is not an all-in answer price: it excludes answer generation, retrieval, storage, retries and any other billed work. The source describes the rates as official OpenAI rates and says its guide was checked November 5, 2026; its example token counts are illustrative, not measured results.

This is a cost guide published by OneMux, which provides access to models through an OpenAI-compatible API. The Sunburst catalogue page lists the model, but the supplied OneMux price snapshot does not establish usable input or output prices for it.

Price evidence and units

The OneMux catalogue snapshot was captured November 14, 2026. The table keeps its unknown values separate from the rates reported by the external source. All token rates below are in US dollars per million tokens; unknown does not mean free.

Billing categoryOfficial rate reported by OfoxOneMux rate in supplied snapshotPublished OneMux price ratio
Text input$5.00Unknown0.11699999999999999
Image input$8.00Unknown0.11699999999999999
Image output$30.00Unknown0.11699999999999999
Text output for a document answerUnknownUnknown0.11699999999999999

The ratio is the model-level value published in the snapshot, not a verified category-specific discount or a price from which to calculate your bill. The snapshot itself marks official input and output prices as unknown; the numerical official rates above come from Ofox's separate guide, not from that snapshot. Confirm current rates and the applicable billing units with your provider before committing funds.

Ofox also reports cached text input at $1.25 and cached image input at $2.00 per million tokens for the official rates. However, its guide does not establish that a particular Sunburst request qualifies for caching. Assume no cache savings until your endpoint and billing records confirm them. OneMux cache prices and tier conditions are unknown in the supplied snapshot; check current OneMux pricing and the applicable model terms rather than extrapolating.

Reproduce an input-only workload estimate

Suppose each question sends a short instruction and one document-page image. These quantities are assumptions for arithmetic, not observed Sunburst usage or a claim that the model supports this request:

AssumptionPer question
Text input120 tokens
Image input1,024 tokens
Image outputNone assumed
Confirmed cache discountNone

Using the rates reported by Ofox:

  • Text input: 120 × $5 ÷ 1,000,000 = $0.0006.
  • Image input: 1,024 × $8 ÷ 1,000,000 = $0.008192.
  • Combined input charge: $0.0006 + $0.008192 = $0.008792 per question.
  • For 1,000 such questions: 1,000 × $0.008792 = $8.792 in input charges.

This is a partial direct-API-style estimate using the source's reported rates, not a OneMux quote. It assumes identical token usage for every question and no image generation. It omits text-answer output because an applicable rate is not supplied. It also omits any separate OCR, embedding, search, storage or model calls. If a different model produces the answer, price that model's inputs and outputs separately.

For a more demanding case, suppose each question uses four page images at the same illustrative 1,024 image tokens per page, with the same 120 text tokens. The image-input component is 4,096 × $8 ÷ 1,000,000 = $0.032768. Adding $0.0006 of text input gives $0.033368 per question, or $33.368 for 1,000 questions, again before all omitted costs. This comparison illustrates how page count affects the estimate; actual image-token usage must be measured.

Make source attribution part of the acceptance test

A cheap response is not useful if a reader cannot trace its claims. Before estimating production volume, define an accepted answer as one that cites the correct document and page or section for each substantive claim, and acknowledges when the supplied material does not support an answer. That is a proposed quality rule, not a documented Sunburst capability.

For a pilot, keep a small set of questions with known source passages, including at least one unanswerable question. For each attempt, record:

  1. The provider, model identifier, endpoint and settings, plus whether the provider documents support for the required image input and text response.
  2. The number of pages sent, text and image usage by category, and any separately billed retrieval or answer-generation calls.
  3. The answer's cited document and page or section, and whether a reviewer could locate the supporting passage.
  4. The actual charge and whether the answer passed the acceptance rule; retain rejected attempts in the sample.

If usage reports show an aggregate input total alongside a category breakdown, do not add both as separate charges. If the breakdown is missing, record the gap rather than assuming every input token was text. If your route is OneMux, integration depends on documented endpoint and parameter support; the supplied material does not establish that this particular workflow has been tested there.

Account for retries without claiming a benchmark

Suppose the hypothetical input-only cost is $0.008792 per attempt and a pilot hypothetically accepts 80% of attempts. To obtain 1,000 accepted answers, the expected attempt count is 1,000 ÷ 0.8 = 1,250. At the same input cost each time, that is $10.99 in input charges. Answer-generation and other costs remain additional. A retry might use different inputs, so replace the constant-cost assumption with measured attempt-level data when available.

When to choose another workflow

If the provider cannot document the required image-to-text response or return enough information to support citations, evaluate a separately documented document-processing and text-answering workflow instead. For documents with extractable text, compare the cost of sending relevant text passages with the cost of sending page images; use each service's actual token counts and applicable rates. Neither route can be declared cheaper or more accurate from the supplied evidence alone.

The practical next step is to verify access and task support, then run a limited pilot with a spending cap. Use its billed usage and citation acceptance rate to replace the illustrative inputs above.

Sources

  • Ofox: GPT Image 2.6 API pricing — source for reported OpenAI token rates, cache cautions and illustrative image-cost arithmetic. The guide says it did not make paid Image 2.6 generation requests.
  • OneMux: GPT Image 2.6 Sunburst — model catalogue listing; the supplied November 14, 2026 snapshot has unknown input and output prices.

Published price snapshot

Captured 2026-11-14T11:30:05.539Z 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 2.6 SunburstUnknownUnknownUnknownUnknown0.11699999999999999

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 2.6 Sunburst cost for document question answering?

An all-in cost cannot be determined from the supplied evidence. At rates reported by Ofox, an illustrative request with 120 text input tokens and 1,024 image input tokens costs $0.008792 for inputs alone. Text-answer generation and other workflow charges are not included.

Does GPT Image 2.6 Sunburst support answers with page citations?

The supplied sources do not establish that capability or identify a supported endpoint for it. Check provider documentation and test citation quality before relying on the model for document question answering.

Can I calculate a OneMux price from the published price ratio?

No. The supplied OneMux snapshot lists input and output prices as unknown. Its published model-level ratio does not establish category-specific rates or an all-in workload price.

Related articles