Hy3 API cost estimates for repeatable evaluations
Estimate Hy3 evaluation costs with explicit token limits, cache assumptions, test cases and reproducible calculations.
Summary · TL;DR
An illustrative 300-request Hy3 evaluation costs $0.057252 in token charges at the OneMux snapshot rates, assuming 3,000 input tokens and 500 output tokens per request with no cache hits. Actual charges depend on billing units, usage, caching and applicable terms.
Key Takeaways
- Budget evaluation runs from request counts and separate input, output and cached-input volumes—not from a blended rate alone.
- The OneMux catalogue snapshot lists a 0.26 price ratio for Hy3; its prices and the external source's prices are separate rate cards and should not be combined.
- Set request and output limits, reserve a retry allowance, and record actual token usage so the estimate can be reproduced.
On this page (9)
The short answer
For a repeatable Hy3 evaluation, estimate each run from the number of requests, the input and output tokens per request, and the expected cache-hit share. Under the illustrative assumptions below, 300 requests cost $0.057252 in token charges at the OneMux snapshot rates, before any other applicable charges. Allowing for up to 10% more requests at the same size raises the planning figure to $0.0629772.
This is a cost estimate, not an evaluation result. OneMux is the publisher and product discussed in this guide; an independent Hy3 pricing page provides a separate rate card. Verify current prices and billing conditions before committing a budget.
Start with a rate card and comparable units
The following OneMux catalogue snapshot was captured on November 17, 2026. The catalogue values supplied for this article do not explicitly state a billing denominator. For the calculations in this guide, they are assumed to be USD per million tokens; confirm that unit and the applicable billing terms on the Hy3 model page before using the results for purchasing decisions. A missing or different unit would change every calculation.
| Hy3 price category | Catalogue official price, assumed per 1M tokens | OneMux price, assumed per 1M tokens |
|---|---|---|
| Standard input | $0.148 | $0.03848 |
| Output | $0.58 | $0.1508 |
| Cache read | $0.035 | $0.0091 |
The published OneMux price ratio is 0.26. It is a catalogue ratio, not a guarantee that a particular workload or invoice will cost that fraction as much. Tier conditions and cache eligibility should be checked against current terms; the supplied snapshot does not specify tiers or cache-write charges. Unknown charges are not zero.
The external LLMLearner Hy3 pricing page, whose rate card is marked effective November 13, 2026, lists $0.132 input, $0.528 output and $0.033 cached input per million tokens. Those are different figures from the OneMux catalogue's official-price column. The supplied information does not explain the difference, so use one rate card consistently and confirm which prices apply to your endpoint. Do not combine one source's input rate with the other source's output rate.
Build a bounded evaluation
Use a fixed test set and define the request shape before running it. This illustrative plan measures repeatability across three runs; it does not claim any Hy3 quality or performance result.
| Planning item | Assumption or limit |
|---|---|
| Test cases | 100 fixed cases |
| Runs | 3 per case, for 300 requests |
| Input size | 3,000 tokens per request, including instructions and test data |
| Output budget | 500 tokens per request |
| Cache reads | None assumed for the baseline |
| Retry allowance | At most 30 additional requests, budgeted at the same size |
The baseline contains 900,000 input tokens and 150,000 output tokens. At the assumed OneMux units:
- Input: 0.9 million × $0.03848 = $0.034632.
- Output: 0.15 million × $0.1508 = $0.02262.
- Total: $0.057252, or $0.00019084 per request.
- With 30 full-size retries: 1.1 × $0.057252 = $0.0629772.
The output figure is a budget, not a prediction that every response will use 500 tokens. Apply an output limit where the endpoint supports one, but still verify how the provider counts and bills output. If input can exceed 3,000 tokens, either reject or separately budget those cases; an output limit alone does not cap the whole request cost. Stop the run when the request or spend budget is reached rather than assuming the estimate enforces a hard billing cap.
Use test cases that expose cost variation
Include short, typical and near-limit prompts in the fixed set. Add cases likely to produce both short and long answers, and record failures or retries separately. Keep the same instructions, case order and output limit across repeat runs; record any settings that can affect responses. If you change the test set or generation settings, label the new run rather than treating it as directly identical.
For a sensitivity check, keep everything else fixed but budget 1,000 output tokens per request. The output volume becomes 0.3 million tokens, or $0.04524; adding the same $0.034632 input charge gives $0.079872 before retries. This is useful because the snapshot output rate is higher than the standard input rate. It does not establish how long Hy3 responses will actually be.
Account for cached input separately
A cache-read price applies only to input actually billed as a cache read. Repeated prompts do not by themselves prove a cache hit, and the supplied information does not establish eligibility, retention or cache-write terms. For a conservative baseline, assume no cache reads; add an alternative scenario only when those conditions are understood.
For example, an assistant workload of 10,000 requests with 8,000 input and 2,000 output tokens each totals 80 million input and 20 million output tokens. If 80% of input is billed as cache reads, the remaining 16 million input tokens use the standard input rate. Using the assumed OneMux units:
| Component | Volume | Rate per 1M tokens | Cost |
|---|---|---|---|
| Standard input | 16M | $0.03848 | $0.61568 |
| Cache reads | 64M | $0.0091 | $0.5824 |
| Output | 20M | $0.1508 | $3.016 |
| Total | 100M | Mixed | $4.21408 |
With no cache reads, the same token volumes cost $6.0944 at those assumed rates. Both totals exclude any other applicable charges. The cache assumption matters more than a single blended price suggests: the total above corresponds to $0.0421408 per million combined tokens, but only for this exact input/output mix and cache share.
Reconcile the estimate after the run
Save the test-set version, request count, output limit, retry count, rate-card date and observed input, output and cache-read token totals. Recalculate using the rates actually applicable to the endpoint, then compare that result with the invoice. Investigate differences in token counts, cache classification, failed-request billing, tiers or additional charges before changing the test plan.
OneMux provides access to models through an OpenAI-compatible API and offers spend visibility. Integration compatibility remains conditional on the documented endpoint and parameter support; this guide does not report a tested OneMux integration. Alternatively, use the rate card and usage records for the provider or endpoint you actually choose.
Conclusion
A useful Hy3 evaluation budget is a reproducible envelope: fixed cases, bounded requests, explicit token assumptions and a separate retry allowance. Treat cache savings as a scenario until usage confirms them, and verify the rate-card units before relying on the dollar totals.
Sources
- LLMLearner: Hy3 pricing, rate card marked effective November 13, 2026; source for the external input, output and cached-input rates.
- OneMux: Hy3 model page, catalogue snapshot captured November 17, 2026; source for the OneMux price values and published ratio used in the illustrative calculations.
Published price snapshot
Captured 2026-11-17T10:30:06.271Z 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 |
|---|---|---|---|---|---|
| Hy3 | $0.148 | $0.58 | $0.03848 | $0.1508 | 0.26 |
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 a 300-request Hy3 evaluation cost?
With 3,000 input tokens and 500 output tokens per request, no cache reads, and the OneMux snapshot prices assumed to be per million tokens, the illustrative token cost is $0.057252. Confirm billing units and applicable charges before relying on that estimate.
Should repeat evaluation runs assume cached input?
Not automatically. Use a no-cache baseline unless the endpoint's cache eligibility and billing conditions are known, then calculate a separate scenario using the expected cache-read share.
Why do the external Hy3 prices differ from the OneMux catalogue?
The supplied sources list different figures but do not explain why. Treat them as separate rate cards and verify the prices that apply to the endpoint you intend to use.
Related articles
Guides
Diagnose Hy3 API Incidents With Cost Signals You Can Reproduce
A practical Hy3 cost-checking guide for API incidents: compare equivalent token units, account for cached input, and use request and failure signals to investigate unexpected spend.
Guides
DeepSeek V4.1 Flash 1017: Estimate API Costs Before an Evaluation
A reproducible guide to estimating DeepSeek V4.1 Flash 1017 API costs, with peak and off-peak rates, cache assumptions, and an evaluation budget.
Guides
DeepSeek V4.1 Flash API costs: a reproducible evaluation budget
Estimate DeepSeek V4.1 Flash evaluation costs with explicit token assumptions, time tiers, cache scenarios, and repeatable test cases.
Guides
GPT 6.2 Sol API cost: a reproducible evaluation budget
Estimate GPT 6.2 Sol evaluation costs with explicit token assumptions, context-tier pricing, cache examples, and repeatable test limits.