Tokenomics
What a million tokens really costs you
UnitVane AI joins billing with token telemetry for every deployment, spreads PTU and reservation cost over the tokens they served, and gives every model, team and cloud one comparable number: effective cost per 1M tokens.
Effective cost per 1M tokens
One number that includes the commitments
List prices only describe pay-as-you-go. UnitVane AI allocates PTU hourly charges and reservations to the deployments and models that used them, then divides by tokens served. The result is what you pay per million tokens, trended week by week.
- Input, output and cached tokens priced separately
- PTU and reservation cost allocated by usage
- Live pay-as-you-go prices from provider price lists
Pricing models
Pay-as-you-go, PTU hourly and reservations, kept apart
Every charge lands in one bucket: pay-as-you-go, PTU billed by the hour, PTU reservation by term, or other. Billing exports often record hourly PTU as pay-as-you-go; UnitVane AI moves those charges to the right bucket in place, without changing daily totals.
- Monthly, 1-year and 3-year reservation terms
- Past months corrected without deleting rows
- Month-close reload with unmatched meters flagged
Model economics
Which workloads could run on a cheaper model
Compare observed cost per 1M tokens across models for the same kind of work, and estimate the monthly saving of moving a workload to a smaller model or to provisioned capacity. Suggestions stay within the same capability tier.
- Savings from observed cost, not list price
- Current model, potential target, token volume
- Guardrails on capability and quality
Token history
Token history that outlasts the provider's retention
Provider metrics expire. UnitVane AI stores daily token and request history per deployment, so last year's cost per 1M tokens is still there when you plan next year's capacity.
- Daily facts per deployment, model and tag
- Historical backfill up to a year from billing
- Summary tables so pages stay fast at fleet scale
Why AI spend needs an owner
What UnitVane AI reads
Billing and telemetry for every provider
| Provider | Cost | Tokens and requests |
|---|---|---|
| Azure OpenAI and Azure AI Foundry | Cost Management, amortized, discovered tenant-wide | Azure Monitor token and request metrics per deployment |
| Amazon Bedrock and SageMaker | CUR 2.0 line items | CloudWatch invocation and token metrics |
| Google Vertex AI | Cloud Billing export in BigQuery | Cloud Monitoring token and request metrics |
| OpenAI and Anthropic APIs | Organization usage and cost APIs | Usage APIs, or the UnitVane AI gateway |
| UnitVane AI gateway | Contracted price per request | Exact tokens per request, key and route |
Tokenomics for every model you run
Cost per 1M tokens
All-in and comparable across models and clouds.
Pricing buckets
PayGo, PTU hourly, reservations and other, separated.
Billing correction
Hourly PTU moved out of pay-as-you-go in place.
Model economics
Savings from moving workloads to a better-priced model.
Long history
Daily token facts kept beyond provider retention.
Owner attribution
Product, team and tenant from tags and gateway keys.
Tokenomics: your questions answered
UnitVane AI
Keep exploring
Evaluate tokenomics with your team
Bring your platform and FinOps priorities. We will walk through the data, controls and evidence for your evaluation.