Agent costs
2026-10-01 to 2026-10-31
LLM cost
$147
2026-10-01 to 2026-10-31
Success rate
86%
492 successful runs
Cost per successful outcome
$0.299
cost per run $0.257
Cache hit rate
77%
cache reads / all input tokens
Gross margin after LLM cost
80%
revenue $746
Cost per day
Cost per day by model
claude-opus-5-5claude-sonnet-5-5gpt-5claude-haiku-5-5
Cost by customer
Model mix (share of cost)
Customers
| Customer | LLM cost | Runs | Cost / success | Revenue | Gross margin |
|---|
| northwind-co | $73 | 226 | $0.380 | $240 | 69% |
| cobalt-labs | $66 | 100 | $0.883 | $60 | -10% |
| bluepeak | $3.30 | 96 | $0.041 | $149 | 98% |
| elmstreet-health | $3.11 | 102 | $0.033 | $149 | 98% |
| dunmore-legal | $1.17 | 21 | $0.056 | $99 | 99% |
| foxglove-retail | $0.03 | 28 | $0.001 | $49 | 100% |
Gross margin here counts LLM cost only. Use the gross-margin-by-customer kit to add hosting and other costs.
Tasks and workflows
| Task | Cost | Runs | Calls / run | Cost / run | Success | Cost / success |
|---|
| research_agent | $132 | 129 | 11.5 | $1.03 | 65% | $1.58 |
| doc_summary | $11 | 195 | 1.5 | $0.058 | 93% | $0.063 |
| code_review | $3.26 | 52 | 2 | $0.063 | 81% | $0.078 |
| support_triage | $0.19 | 197 | 1.6 | $0.001 | 94% | $0.001 |
A failed run still costs money, so cost per success is the number to price and optimize against.
Models
| Model | Cost | Share | Calls | Input tokens | Output tokens | Cache hit |
|---|
| claude-opus-5-5 | $132 | 90% | 969 | 95,148,382 | 2,408,955 | 84% |
| claude-sonnet-5-5 | $11 | 8% | 294 | 5,703,213 | 268,905 | 28% |
| gpt-5 | $3.26 | 2% | 102 | 1,639,670 | 178,486 | 31% |
| claude-haiku-5-5 | $0.70 | 0% | 820 | 4,765,989 | 594,915 | 18% |
Input tokens include cache reads and writes. Cache hit = cache reads / input tokens.
Source: examples/usage.csv. Blank costs priced from examples/prices.csv; check those prices against each provider's current pricing.