03 / Model the cost
Choose a model. Show your math.
Compare Claude models for an agency-owned application, then estimate direct API token charges from your expected workload. Start with a bounded pilot and test model quality on your own tasks.
Explore government use cases Jump to API cost estimateCompare / Claude API
Four models to evaluate
These are starting hypotheses, not a quality ranking for government work. Prices are standard base rates per million tokens (MTok).
Model specifications and descriptions: Anthropic model overview ↗. Availability, features, and rates can change.
Estimate / Direct API
Put your workload into the estimate
Enter the average tokens consumed per request and the number of requests per month. The same workload is applied to every model below.
- Input tokens / month
- 12.5M
- Input charge
- $25.00
- Output tokens / month
- 3M
- Output charge
- $30.00
- Annual run rate
- $660.00
(input tokens × requests ÷ 1,000,000 × input $/MTok) + (output tokens × requests ÷ 1,000,000 × output $/MTok)
Same workload, different base rates
Cost is one consideration. Test answer quality, latency, escalation, and human review effort before choosing a model.
| Model | Input / MTok | Output / MTok | Monthly estimate |
|---|
Before budgeting
What the token estimate leaves out
Integration, hosting, retrieval, logging, evaluations, security review, support, and staff time.
Prompt caching, Batch API, tools, long-context or fast modes, data residency, partner platforms, taxes, and negotiated terms.
Retries, tool calls, changing prompt length, output variation, and model-specific tokenization. Measure actual usage and review invoices.