Understanding cost & value
Cost & Value keeps three questions separate:
- What did you commit to pay for the plan?
- What cost belongs to the selected analytics range?
- What would the observed workload cost at maintained API list prices?
Terms
| Term | Meaning |
|---|---|
| Subscription commitment | The configured plan price for its full cadence, such as USD 20/month. |
| Subscription allocation | The share of that commitment overlapping the selected range. The economics API returns this as cost_basis. |
| PAYG cost | Spend for a pay-as-you-go provider, from one authoritative or safe fallback source. |
| Provider-reported cost | A monetary counter/delta returned by the provider. |
| Provider billing history | Dated monetary buckets returned by a native billing/cost endpoint. |
| Pricing estimate | Reconstructed cost from model, token class, timestamp, and the maintained pricing catalogue. |
| Hermes-observed cost/workload | Cost metadata, tokens, requests, and sessions seen by Hermes; it may cover only part of account activity. |
| API-equivalent value | List-price estimate for the Hermes-observed model/token workload. |
| Tokens/$ | Observed attributed tokens divided by the selected cost basis. |
| Requests/$ | Observed attributed requests divided by the selected cost basis. |
| Cost/1M tokens | Selected cost basis divided by attributed tokens, scaled to one million. |
| Value multiplier | API-equivalent value divided by cost basis. |
PAYG source precedence
The dashboard selects exactly one cost representation:
- provider-reported spend;
- provider-native billing history;
- pricing estimate with at least 80% token pricing coverage;
- unavailable.
The selected representations are never added or averaged. This prevents a provider's rolling spend, billing buckets, and reconstructed price from counting the same purchase multiple times. When actual spend and a reconstruction both exist, actual spend is authoritative and the reconstruction is corroborating; a material disagreement is returned in analytics API reconciliation metadata.
An estimate at 80–94.99% coverage is marked partial but may be used. Below 80%, it remains diagnostic API-equivalent data and cannot be the PAYG cost basis or a whole-workload efficiency denominator. Unknown or unpriced tokens are never silently assigned zero cost.
Subscription allocation and the current UI
Subscription cost is allocated by the real overlap between the selected range and billing periods. Month lengths, leap years, monthly/yearly cadence, and an optional billing anchor are respected. Without an anchor, allocation is aligned to the selected range and explicitly marked estimated.
The configured USD 20/month remains the commitment; the economics API calculates only the overlapping portion as its selected-range cost_basis. The current Usage Overview and Cost & Value table display the full commitment and use it for their subscription efficiency ratios. The allocated value is currently API detail rather than a separate value in the table.
Examples
| Connection | Billing configuration | Interpretation |
|---|---|---|
| OpenAI Codex | Subscription, USD 20/month | Display the USD 20 commitment; the API also computes selected-range allocation. Compare only mapped Hermes-observed workload. |
| OpenCode Go | Subscription, USD 10/month | Treat quota-window dollar figures as allowance, not PAYG billed spend. The UI displays the USD 10 commitment. |
| DeepSeek | PAYG | Prefer provider monetary spend when available. The standard balance endpoint is not spend history, so sufficiently covered Hermes token pricing can provide an estimated cost basis. |
Free connections use a zero configured cost basis. When required billing, attribution, pricing, or compatible currency data is absent, efficiency is unavailable instead of fabricated. No foreign-exchange conversion is performed.
See Billing configuration and Cost accounting.