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.
Summary · TL;DR
Estimate document question-answering costs by separating uncached input, output, cache reads and writes, then multiplying each token count by the matching published rate. A clearly labeled example shows the calculation for a source-attributed answer.
Key Takeaways
- A token estimate is only as useful as its workload assumptions: count the document context, question, instructions, and generated answer separately.
- For an illustrative 12,000 input-token and 1,000 output-token request, the listed OneMux rates calculate to $0.038896; actual usage and eligible billing conditions may differ.
- The supplied OneMux catalogue does not list Claude Fable 5.2 cache prices, so cache-related OneMux costs cannot be calculated from this snapshot.
- Judge document Q&A by whether answers are supported by identifiable passages, not just by token cost.
On this page (7)
Estimate the workload before estimating the bill
For document question answering with source attribution, estimate one complete request: the instructions, question, document text sent to the model, and the answer with its citations or references. Count the input and output separately. Do not assume that a document’s file size equals its token count, or that a short question produces a short answer.
The calculation below is an illustrative budget, not a measured result or a prediction of Claude Fable 5.2 usage. It assumes one synchronous request, no retries, no separately billed tools, 12,000 input tokens, and 1,000 output tokens. The input includes the question, instructions, and supplied document context; the output includes the answer and source references.
Published prices and units
The OneMux catalogue snapshot was captured on November 15, 2026. The standard prices below are stated in USD per million tokens. Official prices are the catalogue’s listed official rates; OneMux prices and the published ratio are reproduced from that same snapshot. This is a price comparison, not a benchmark, invoice, or promise of savings.
| Price type | Input per million tokens | Output per million tokens |
|---|---|---|
| Official listed price | $8 | $40 |
| OneMux published price | $2.288 | $11.44 |
| Published OneMux-to-official price ratio | 0.286 | 0.286 |
The ratio is the catalogue’s published value, not a percentage discount rounded or recalculated here. Rates, billing units, and eligibility can change; check the Claude Fable 5.2 model page and OneMux pricing before budgeting. The model page lists a 1M context value, but that does not imply every document fits in a request: instructions and conversation history also consume context.
Reproduce the example calculation
For each token category, multiply the token count by its per-million rate and divide by 1,000,000. Add input and output charges.
| Category | Assumed tokens | Official-rate arithmetic | OneMux-rate arithmetic |
|---|---|---|---|
| Input | 12,000 | 12,000 × $8 ÷ 1,000,000 = $0.096 | 12,000 × $2.288 ÷ 1,000,000 = $0.027456 |
| Output | 1,000 | 1,000 × $40 ÷ 1,000,000 = $0.040 | 1,000 × $11.44 ÷ 1,000,000 = $0.01144 |
| Illustrative total | 13,000 | $0.136 | $0.038896 |
The total is per request under those assumptions. A workload estimate needs the number of requests too: for 1,000 identical requests, the arithmetic gives $136 at the listed official rates or $38.896 at the listed OneMux rates. That multiplication assumes the same token counts and applicable rates on every request. It does not account for retries, cache billing, tools, or other charges.
Treat cache prices as a separate unknown
The supplied OneMux catalogue snapshot does not provide Claude Fable 5.2 cache-read or cache-write prices. Do not substitute another model’s cache rates or count cached input as free. If your implementation uses prompt caching, identify cache writes, cache reads, uncached input, and output in usage records, then use the applicable provider or route pricing before calculating a total.
A dated comparison published November 10, 2026, reports official Fable 5.2 cache-read pricing of $0.20 per million tokens, five-minute cache writes at $10 per million, and one-hour cache writes at $16 per million. These are separate official rates from the standard input and output prices above, not OneMux cache prices. The article notes that cache creation has its own charge and that a cache directive alone does not make every subsequent request a cache hit. See the dated comparison and its cited model sources and confirm current applicable pricing before using these figures.
Make attribution part of acceptance
A cost estimate does not tell you whether an answer is grounded in the documents. For a useful document-Q&A comparison, define acceptance before running requests:
- Each material factual claim must point to a source the reviewer can locate, such as a document title and page or section.
- The cited passage must support the claim, rather than merely mention the same topic.
- Missing or conflicting evidence must be surfaced instead of filled with an unsupported answer.
- Measure input and output tokens alongside accepted answers, incomplete responses, refusals, and retries.
These checks keep a fluent but unsupported response from counting as a successful, low-cost task. They also make token budgets more realistic: requiring traceable evidence may change the amount of context supplied and the length of the requested answer. No success rate or document-Q&A performance result is established by the supplied price data.
Turn the estimate into a workload budget
Start with representative documents and questions, then record actual input and output usage for each request. Include the full context your production workflow sends, not just the question. Keep retries and failed or rejected answers in the totals. For each billing category with a known rate, multiply usage by that rate using matching units; leave categories with unknown prices marked unknown until verified.
OneMux publishes access to models through an OpenAI-compatible API and focuses on model access, routing, keys, spend visibility, and pay-as-you-go usage. Compatibility with a specific model and request parameter depends on documented endpoint and parameter support; do not assume every provider-specific feature, including caching, behaves identically through every route.
A defensible estimate is a transparent budget, not a performance claim: state token counts, rates, cache assumptions, request volume, and what counts as an accepted answer. Recalculate when the workload or published prices change.
Published price snapshot
Captured 2026-11-15T13:00:05.811Z 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 |
|---|---|---|---|---|---|
| Claude Fable 5.2 | 8 | 40 | $2.288 | $11.44 | 0.286 |
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 Claude Fable 5.2 document-Q&A request cost?
It depends on input and output token counts and applicable billing categories. In the illustrative 12,000-input and 1,000-output example, the arithmetic is $0.136 at the listed official rates and $0.038896 at the OneMux catalogue rates, before cache charges, retries, or other costs.
Are Claude Fable 5.2 cache prices included in the OneMux estimate?
No. The supplied OneMux price snapshot does not list Fable 5.2 cache prices. The dated external comparison reports official cache rates, but those should not be treated as OneMux rates.
How should I count tokens for document question answering?
Count all text sent as input, including instructions, the question, relevant document context, and conversation history. Count the generated answer and its source references as output. Use actual request usage when available rather than treating file size as token count.
What makes a source-attributed answer acceptable?
Set a rule that each material claim has a locatable citation and that the cited passage supports that claim. Track unsupported answers, conflicting evidence, incomplete responses, and retries rather than counting only successful-looking answers.
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
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
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.
Guides
Gemini 3 Pro Image Preview API costs for document question answering
Estimate document question-answering costs using explicit token assumptions, comparable price units, and a reproducible monthly calculation.