Usage & Spend
Use Usage and Credits views in the OpenModels dashboard to understand how requests consume tokens and credits.
What to monitor
Monitor these signals during development and production:
API key breakdown
Create separate API keys for environments and workloads when you need clean usage attribution.
Examples:
developmentstagingproductionsupport-agentbatch-jobs
This makes it easier to identify which workload consumed credits and which monthly limit should be adjusted.
Model breakdown
Use model-level spend to compare cost and quality. If a workload is expensive, check whether:
- Prompts are larger than expected
- Outputs are too long
- A high-cost model is used where a lower-cost model would work
- Retries are increasing request volume
- Tool calling adds extra messages to the conversation
Cost investigation workflow
When spend looks higher than expected:
- Check the time window where spend increased.
- Break down usage by API key.
- Break down usage by model.
- Compare input tokens and output tokens.
- Check whether retries or long outputs increased token volume.
- Adjust model choice,
max_tokens, prompts, or monthly spend limits.
Production alerts
For production workloads, use monthly spend limits and auto top-up together:
- Monthly spend limits prevent one key from overspending.
- Auto top-up helps prevent balance interruptions.
- Usage monitoring shows which key or model is driving spend.
If requests fail with insufficient balance or spend limit errors, review Credits, Usage, and Rate Limits together.