Coming soon Estimates API
Know what agent work will cost before it starts.
Describe the work. Get back the chance of every cost in dollars, learned from what similar work really cost.
Check your inbox
We sent a link to . Confirm it, and you hear from us first when estimates open.
RequestPOST /api/v1/estimates { "work": "Refund checkout for a returns portal", "model": "claude-opus-5-5", "context": "Last flow: 702.40 USD, sandbox timeouts" }
Answer200 OK { "id": "est_4k2q9w", "currency": "USD", "median": "610.00", "distribution": [ { "up_to": "250.00", "p": 0.04 }, { "up_to": "500.00", "p": 0.31 }, { "up_to": "750.00", "p": 0.41 }, { "up_to": "1000.00", "p": 0.17 }, { "up_to": null, "p": 0.07 } ] }
Ask before you spend.
Client quotes
Quote a new project with a range you can defend.
Agents that ask first
Before a big task, an agent asks what it will cost and picks the model to match.
Fixed prices
Price AI work by the result and know your margin before you say yes.
Every estimate gets checked.
- 1
Estimate
Ask before the work starts.
- 2
Record
Agents report what they spent and sum up what they built.
- 3
Match
Our classifier puts each report next to its estimate. The delta shows scope creep and trouble.
- 4
Learn
Every match trains the next estimate.
Built on real work.
Agents misjudge their own cost. In a 2026 study, their guesses of their own token use matched the real use with a correlation of 0.39 at best, and they guessed too low. Read the study
Pooled real costs
Every organization's token entries, matched to provider bills, train one classifier. It answers with probabilities only, so no description, name or client comes back out.
Public benchmark
Every month a fixed set of tasks runs through the API. We publish how close the estimates land, next to a plain LLM's guess.
Pay per estimate
One price per estimate, however much context you send. More context, better estimates.
Your own history makes the best estimates.
Start recording now. Solo is free forever and needs no card.