GPT Image 2.6 Sunburst Cost Estimation for Support Triage With Human Approval
A reproducible way to estimate GPT Image 2.6 Sunburst workload costs for customer support triage, including retries, reference images, human approval, and unknown OneMux catalogue prices.
Summary · TL;DR
For support triage, the useful budget is the cost per approved case, not the nominal cost of one successful API response. Keep text input, image input, image output, cache treatment, retries, and any additional model work separate. The supplied November 5, 2026 pricing source lists the same GPT Image 2.6 token rates for Sunburst and Flare, but it does not establish identical token consumption or a universal cost for GIF and SVG work. OneMux lists GPT Image 2.6 Sunburst as available through unified model routing, while its November 16, 2026 catalogue snapshot leaves the model's input and output prices unknown.
Key Takeaways
- Use the documented Sunburst rates of $5 per million text input tokens, $8 per million image input tokens, and $30 per million image output tokens as a starting assumption, not a measured workload price.
- Estimate accepted support cases by dividing the number of desired approvals by the observed acceptance rate, then multiplying by the measured cost per attempt.
- Do not treat unknown OneMux input or output prices as zero, and do not derive a dollar price from the catalogue's published ratio alone.
- Human approval should be part of the acceptance rule: correct diagnosis, usable response draft, accurate visual details, and no unsafe or misleading recommendation.
- GIF and SVG workflows require separate usage records because frame work and vector authoring can add costs that a still-image estimate does not capture.
On this page (10)
The decision to make first
For customer support triage, estimate the cost of an approved case, not merely the cost of an API response. A response can return successfully and still be rejected by an agent because it misread a screenshot, missed a product detail, or proposed an unsuitable reply.
A practical workflow is:
- A customer submits text and, where relevant, a screenshot or product image.
- GPT Image 2.6 Sunburst produces a structured visual assessment or supporting image output.
- A support agent reviews the result and approves, edits, or rejects it.
- Only approved cases continue to the customer-facing workflow.
OneMux is the publisher of this guide and provides access to leading AI models through one OpenAI-compatible API, with model routing, key management, spend visibility, credit top-ups, and pay-as-you-go usage. Compatibility with a particular request parameter or response field remains conditional on the documented endpoint support. This workflow has not been tested through OneMux, and the examples below are cost-planning examples rather than paid benchmark results.
Published Sunburst rates and units
The supplied source, checked November 5, 2026, reports these official OpenAI rates for gpt-image-2.6-sunburst and gpt-image-2.6-flare:
| Billing category | GPT Image 2.6 Sunburst | Unit |
|---|---|---|
| Text input | $5.00 | 1 million tokens |
| Cached text input | $1.25 | 1 million tokens |
| Image input | $8.00 | 1 million tokens |
| Cached image input | $2.00 | 1 million tokens |
| Image output | $30.00 | 1 million tokens |
A listed cache rate is not proof that a request qualifies for it. The source says the current image guide explicitly names earlier Image 2 and 2.5 models for cache-related eligibility; it does not extend that statement to 2.6. Exclude cache savings from a forecast until your route and billing record confirm them.
The source also does not show that Sunburst and Flare consume equal tokens for the same request. Model choice, image dimensions, quality, input material, and attempts still affect workload cost.
Reproducible cost calculation
For one request without a confirmed cache discount and without an additional model charge, calculate:
Cost = (text input tokens x 5 + image input tokens x 8 + image output tokens x 30) / 1,000,000
Illustrative example:
Text input: 120 tokens
Image input: 0 tokens
Image output: 1,000 tokens
Cost = (120 x 5 + 0 x 8 + 1,000 x 30) / 1,000,000
= 30,600 / 1,000,000
= $0.0306 per attempt
If the case includes 1,024 image-input tokens, add $0.008192:
(120 x 5 + 1,024 x 8 + 1,000 x 30) / 1,000,000 = $0.038792
These token counts are assumptions for arithmetic. They are not a prediction of Sunburst consumption. Keep text and image inputs separate, and do not bill both an aggregate input total and its detailed components. If usage details are missing, record the missing information instead of silently classifying it as text input.
Convert attempts into approved cases
Suppose a pilot records an average cost of $0.0306 per attempt and agents approve 80% of results. To obtain 100 approved cases:
Expected attempts = 100 / 0.80 = 125
Expected generation cost = 125 x $0.0306 = $3.825
At a 50% approval rate, the same target requires 200 attempts:
200 x $0.0306 = $6.12
The approval rule should be explicit. For support triage, it might require that the visual observation is accurate, the relevant product or UI detail is preserved, the proposed category is appropriate, and a human agent signs off before any customer-facing action. Record rejected outputs and their reasons. An HTTP 200 response is not an approval and is not evidence that the result met the support task.
For each pilot request, save the model identifier, endpoint, image size, quality, text and image input counts, output usage, actual charge, elapsed time, approval decision, and rejection reason. This makes the next estimate reproducible.
OneMux price snapshot: what is known
OneMux's catalogue snapshot was captured on November 16, 2026. It lists GPT Image 2.6 Sunburst as available through unified model routing, but the published input and output prices are unknown:
| Model | Official input | Official output | OneMux input | OneMux output | Published price ratio |
|---|---|---|---|---|---|
| GPT Image 2.6 Sunburst | Unknown | Unknown | Unknown | Unknown | 0.11699999999999999 |
Unknown values are not zero. The ratio is reproduced exactly from the supplied catalogue snapshot; because the corresponding dollar prices and billing conditions are unknown, it cannot be converted here into a OneMux cost or described as a guaranteed discount. Check the Sunburst model page and OneMux pricing for current published terms before funding a workload.
Do not compare providers unless the input and output units match. Tiered, cached, or time-based conditions can change the calculation. Keep the provider, endpoint, request settings, returned usage, and settled account charge in the same record.
GIF and SVG are different workloads
A still-image estimate is not a universal price for other deliverables. GIF work can involve multiple frames, dimensions, duration, and loop acceptance. Do not multiply a still-image cost by frame count unless the provider documents that as its billing method. Record returned frames, playback duration, dimensions, usage, and charge.
SVG work may include reference-image generation plus a separate authoring-model charge. The supplied source does not name a universal authoring-model rate, so do not quote an all-in SVG price by applying the $30 image-output rate to every component. Record each billed component independently and inspect the SVG for usable vector elements before approval.
Integration prerequisites and failure checks
Before implementing the triage worker, confirm that the chosen endpoint supports the Sunburst model identifier, image inputs, the desired output format, usage reporting, and the request controls you plan to use. Keep API credentials server-side and use a small spending limit for the first sample.
A conditional OpenAI-compatible request might look like this, subject to the provider's documented schema:
curl https://YOUR_DOCUMENTED_ENDPOINT/v1/responses \
-H "Authorization: Bearer $API_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "gpt-image-2-6-sunburst",
"input": "Inspect the attached support screenshot. Identify the visible issue and list evidence for a human agent to review. Do not send a customer-facing reply."
}'
If the endpoint rejects the model, reports no usage, or returns an input breakdown you cannot reconcile, stop the estimate rather than substituting a different model or treating missing fields as zero. For an existing 2.5 workflow, treat migration to 2.6 as a new regression check.
Conclusion
Budget Sunburst support triage around accepted, human-approved cases. Start with explicit token assumptions, measure real usage and approval rates in a small pilot, then replace assumptions with settled charges. The official rates provide comparable arithmetic; they do not remove uncertainty about consumption, cache eligibility, retries, or additional authoring work. OneMux can provide a unified routing path, but the current supplied catalogue snapshot does not publish Sunburst input or output dollar prices, so a OneMux-specific estimate requires current documented rates or measured billing data.
Sources
- GPT Image 2.6 API pricing: image costs and GIF/SVG billing — supplied source for the November 5, 2026 official-rate summary, calculation examples, cache caveats, and GIF/SVG billing considerations.
- GPT Image 2.6 Sunburst on OneMux — supplied catalogue reference.
- OneMux pricing — current pricing reference to check before committing a workload.
Published price snapshot
Captured 2026-11-16T12:30:05.587Z 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.
| Model | Official input | Official output | OneMux input | OneMux output | Published price ratio |
|---|---|---|---|---|---|
| GPT Image 2.6 Sunburst | Unknown | Unknown | Unknown | Unknown | 0.11699999999999999 |
Current OneMux prices. Compare workload costs only when input/output and billing units match; see each model page for tier conditions.
FAQ
What is the GPT Image 2.6 Sunburst price per support triage request?
There is no single reliable per-request price. Under the supplied official rates, a hypothetical request with 120 text input tokens and 1,000 image output tokens costs $0.0306 before image input, cache, retries, or additional model work. Measure actual usage for your support prompts and images.
How should I budget for human approval?
Divide the number of desired approved cases by the observed approval rate to estimate attempts, then multiply by measured cost per attempt. For example, 100 approvals at an 80% approval rate require an expected 125 attempts.
Are OneMux GPT Image 2.6 Sunburst prices published?
The supplied November 16, 2026 OneMux catalogue snapshot marks official input, official output, OneMux input, and OneMux output prices as unknown. It records a published ratio of 0.11699999999999999, but that ratio cannot be converted into a dollar estimate without matching published prices and billing units.
Can I use the still-image calculation for GIF or SVG support assets?
Not without further evidence. GIF jobs may involve frame and duration-related workload, while SVG jobs may include reference-image generation and an authoring-model charge. Record each billed component and use provider billing data.
Related articles
Guides
Gemini 4 Carbon API Cost: A Reproducible Support-Triage Estimate
Estimate a Gemini 4 Carbon customer-support triage task with explicit token assumptions, cache accounting, human approval, retries, and OneMux’s published prices.
Guides
DeepSeek V4.1 Flash 1017 Cost Guide for Human-Approved Support Triage
Estimate DeepSeek V4.1 Flash 1017 API costs for customer support triage with explicit token assumptions, peak and off-peak pricing, caching, and reproducible monthly calculations.
Guides
GPT 6.2 Astra API cost estimate for support triage with human approval
Estimate GPT 6.2 Astra token costs for a human-approved customer support triage workflow, with tier and cache examples.
Guides
GPT Image 2.6 API Cost Estimation for Privacy-Aware Workloads
Estimate GPT Image 2.6 workload costs using explicit token assumptions, acceptance rates, and privacy-conscious measurement practices.