Estimate GPT 6.1 Terra API Costs for Document Q&A
A reproducible way to estimate GPT 6.1 Terra costs for document question answering with source attribution, including cache assumptions and the long-context tier.
Summary · TL;DR
Estimate document question-answering costs by counting uncached input, cached input, cache writes and output separately. This guide works through an illustrative source-attributed Q&A workload using GPT 6.1 Terra’s published rates and makes the assumptions and arithmetic explicit.
Key Takeaways
- For a request under the 272K input-token threshold, estimate input and output separately; the illustrative 24,000-input, 700-output request costs $0.004602 at the listed OneMux rates, before any cache charges.
- Source attribution affects cost when it increases the prompt or answer token count. The model’s ability to cite sources is not a guarantee that citations are correct.
- Do not assume caching saves money: include cache-write charges and only count cache reads when the same content is reused.
- Requests above 272K input tokens use a higher rate tier in the supplied pricing data; recalculate the whole request at that tier.
On this page (9)
Document question answering costs are easiest to underestimate when the estimate counts only the document text. A useful budget also includes the question, instructions, retrieved passages, any source labels or citation format, and the model’s answer. This guide shows how to estimate those pieces for GPT 6.1 Terra, with a worked example you can replace with your own token counts.
OneMux is the publisher and product discussed here. Prices below are a published catalogue snapshot, not a live benchmark. The example is illustrative, not a measurement of a deployed workload.
The rates and what they cover
The OneMux catalogue snapshot was captured on 2026-11-08. The listed prices are per million tokens. The official rates and OneMux rates are shown as published; the catalogue’s discount ratio is 0.156 for the listed input and output rates. That ratio means the OneMux listed rate is 15.6% of the corresponding official rate, or 84.4% lower for those listed rates. It does not establish a discount on every request: tier, cache, and other billing conditions matter.
| GPT 6.1 Terra rate | Official price per million tokens | OneMux price per million tokens |
|---|---|---|
| Input, up to and including 272K input tokens | $1.00 | $0.156 |
| Output, up to and including 272K input tokens | $5.00 | $0.78 |
| Cached input, up to and including 272K | $0.05 | $0.0078 |
| Cache write, standard duration | $1.25 | $0.195 |
| Cache write, 12-hour duration | $2.50 | $0.39 |
| Input, above 272K input tokens | $2.00 | $0.312 |
| Output, above 272K input tokens | $7.50 | $1.17 |
| Cached input, above 272K | $0.10 | $0.0156 |
| Cache write, above 272K, standard duration | $2.50 | $0.39 |
| Cache write, above 272K, 12-hour duration | $5.00 | $0.78 |
The tier boundary matters: according to the supplied pricing reference, requests above 272K input tokens are charged at the higher tier for the request, not just for tokens beyond the boundary. Check the model’s current OneMux listing before budgeting; rates and billing conditions can change.
A repeatable estimate for source-attributed Q&A
For each request, record four token counts separately:
- Uncached input: question, document text or retrieved passages, instructions, and source labels sent to the model.
- Cached input: previously cached content actually read for this request.
- Cache writes: content newly written to cache, if applicable.
- Output: the answer, including requested quotations, source IDs, or citation details.
Then calculate each line using its matching per-million-token price. For example, at the OneMux standard rates below the threshold:
- Uncached input cost = input tokens ÷ 1,000,000 × $0.156
- Cached input cost = cached input tokens ÷ 1,000,000 × $0.0078
- Cache-write cost = cache-write tokens ÷ 1,000,000 × $0.195
- Output cost = output tokens ÷ 1,000,000 × $0.78
Add the line items. Do not apply the cached-input rate to new content being written, or count the same tokens as both cached and uncached input.
Worked example: ask questions across a document set
Assume one request sends 24,000 uncached input tokens: 20,000 tokens of selected document passages, 1,000 tokens of question and instructions, and 3,000 tokens for source labels and context. Assume the answer uses 700 output tokens to give a response with source references. The request is below the 272K threshold, uses no cache, and has no retries.
- Input: 24,000 ÷ 1,000,000 × $0.156 = $0.003744
- Output: 700 ÷ 1,000,000 × $0.78 = $0.000546
- Total: $0.004290 per request
At 10,000 requests with the same token counts, the estimate is $42.90. This is a simple multiplication of the illustrative per-request cost, not a forecast of actual usage. It excludes retries, other services, and any applicable fees not represented in the listed token prices.
One way to make the estimate more realistic is to use counts from a representative sample of your own questions and retrieved passages. Track typical and high-end input sizes separately: a small question that retrieves two passages has a different cost from one that retrieves a dozen. Include the citation format you intend to ship, because quoted text and source identifiers can increase both input and output tokens.
When caching helps—and when it adds cost
Caching is most relevant when the same content is reused across requests. A one-off document sent as a new cache write is not automatically cheaper than uncached input. At the listed standard OneMux rates below 272K, writing one million tokens costs $0.195; reading one million cached tokens costs $0.0078, compared with $0.156 per million uncached input tokens.
For an illustrative one-million-token cache write followed by reads of that same content, the cache-write line alone equals the cost of 1.25 million uncached input tokens at the listed rates. The actual break-even depends on cache behavior and which content is reused. Estimate cache writes and reads separately, and do not assume the cache stays available or is reused unless your implementation and provider documentation confirm that behavior.
The long-context cost check
If a request exceeds 272K input tokens, use the higher-tier rates for the whole request. For example, an illustrative request with 300,000 uncached input tokens and 3,000 output tokens costs:
- Input: 300,000 ÷ 1,000,000 × $0.312 = $0.0936
- Output: 3,000 ÷ 1,000,000 × $1.17 = $0.00351
- Total: $0.09711
That calculation uses the OneMux rates in the supplied snapshot. For comparison, using the below-threshold rates for the same token counts would produce $0.04914. That lower figure is not the correct estimate for a request above the threshold under the supplied tier rules.
For document Q&A, a practical response to unexpectedly large prompts is to inspect retrieval and prompt construction: are irrelevant passages being included, or are repeated instructions taking up space? Reducing input tokens can lower cost, but keep enough evidence for the answer and its citations. Do not trade away the source context needed to verify a response just to hit a token target.
A budgeting checklist
- Measure input and output tokens on representative questions, including citation formatting.
- Separate uncached input, cached reads, and cache writes in the estimate.
- Check whether any request exceeds 272K input tokens and apply the appropriate tier.
- Multiply by expected request volume, then add a separate allowance based on observed retries.
- Validate a sample of answers and source references. Token prices do not establish answer quality or citation accuracy.
- Recheck published prices before committing to a budget; the OneMux snapshot is dated, not live.
For current model availability and the listed rates, consult the GPT 6.1 Terra model page. OneMux provides access to models through an OpenAI-compatible API, but compatibility with a particular endpoint or parameter depends on documented support. This guide does not claim that an integration or workload has been tested.
Bottom line
A defensible GPT 6.1 Terra estimate starts with your own token counts, applies the matching input, output, cache, and context-tier rates, and states what the arithmetic excludes. For document Q&A with source attribution, count the evidence and citation format as part of the workload—not as free extras.
Sources
- GPT 6.1 Terra pricing reference — supplied source for the model’s pricing structure and the 272K tier behavior.
- OneMux GPT 6.1 Terra listing — published catalogue rates used here, snapshot dated 2026-11-08.
Published price snapshot
Captured 2026-11-08T13:00:05.237Z 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 6.1 Terra | $1 | $5 | $0.156 | $0.78 | 0.156 |
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 one GPT 6.1 Terra document Q&A request cost?
It depends on token counts and billing tier. In the illustrative example here, 24,000 uncached input tokens and 700 output tokens cost $0.004290 at the listed OneMux rates below the 272K threshold, excluding retries and other charges.
Are source citations included in the input or output token estimate?
Count source labels and passages sent to the model as input, and citation text returned in the answer as output. Their token counts depend on your prompt and response format.
Does caching always reduce GPT 6.1 Terra costs?
No. A cache write has a separate charge, while cached reads use a lower listed rate. Caching is potentially useful when content is reused; a one-off write may not pay back. Model cache behavior and retention should be checked against current provider documentation.
What happens when a request exceeds 272K input tokens?
The supplied pricing reference describes a higher rate tier that applies to the whole request. Recalculate all applicable input, output, and cache lines using that tier rather than charging only excess tokens at the higher rates.
Does this cost estimate prove GPT 6.1 Terra will produce accurate citations?
No. Pricing data describes token charges, not answer quality or citation accuracy. Test source attribution against your own documents and evaluation criteria.
Related articles
Guides
GPT 6.2 Astra API cost estimate for document question answering
Estimate GPT 6.2 Astra document question-answering costs using explicit token assumptions, context tiers, cache rates, and reproducible arithmetic.
Guides
Estimate Claude Fable 5.2 API Costs for Document Q&A
A reproducible way to estimate Claude Fable 5.2 costs for document question answering, with explicit token assumptions, source-attribution guidance, and dated price figures.
Guides
Qwen3.7 Plus API cost for document question answering
Estimate Qwen3.7 Plus API costs for document question answering with source attribution, using explicit token assumptions and reproducible arithmetic.
Guides
GPT 6.1 Terra API Cost: A Reproducible Cache Evaluation
Estimate GPT 6.1 Terra API costs with explicit token assumptions, published OneMux and official prices, and a repeatable test for whether prompt caching helps your workload.