A base fee, the compute you reserve, the seconds you use.
Prepaid credits, spent per second of the compute a solve reserves. You pick the engine, the compute preset it runs on and the runtime; the maximum cost follows from those three and is known before you submit.
How a solve is priced
cost = base_fee + ceil(seconds × effective_rate)
Three things set the price. The engine's base fee, charged once per solve. Its runtime rate, in credits per second of reserved compute. And the compute preset you run on, which multiplies that rate: the effective rate is the engine's rate times the preset's multiplier. The rate card below prints the effective rate for every engine on every preset it offers.
The maximum cost is known before you submit
Take the effective rate, multiply by the runtime you set, round up, add the base fee. That is the most the solve can cost, and it is what is reserved from your balance when the problem is admitted. You pay the seconds you used, never more. Queue time is never billed, a solve that never starts costs nothing, and a cancelled one costs the seconds it ran.
A solve that runs s seconds on an engine with base fee b at an effective rate r costs b + ceil(s × r) credits.
The terms you submit at are the terms you pay: the base fee, the rate and the preset's multiplier are captured with your submission, so a later price change never touches a solve in flight. The meter is the time your solve held the compute, measured on our clock from the moment an engine picks the problem up to the moment it ends and capped at the runtime you set. It comes back with the solution and it is itemised on your ledger.
| Engine | Slug | Base fee | CPU VPS | Status |
|---|---|---|---|---|
| per solve | 4 vCPU, 8 GB memory | |||
| highs-mof | 0 credits | 0 credits default | Answering |
Values come from the live catalogue. Every figure under a preset is that engine's rate on that compute, per second. A hyphen means the engine does not run on that preset. A meta-solver has no price of its own: you pay the engine that solved, at its own base fee and rate, which is why its figures are ranges. The catalogue names the members each one routes across. Commercial engines run under a licence we hold, priced into the rate.
Cost at 6 s, in credits
- HiGHS (MathOptFormat) highs-mof 0 to
An engine that does not offer the selected preset is dimmed here and drawn nowhere.
Move across the chart to read a second, and click to keep it. Near a curve, that curve raises and names the engine it belongs to. Choose engines in the list to plot only those, and the scale follows them.
HiGHS (MathOptFormat)
- Base fee
- 0 credits
- Effective rate
- 0 credits per second
- Cost at 6 s
- 0 to credits
- Maximum cost at 12 s
- 0 to credits
ceil() rounds every solve up to the next whole credit. Each curve is one engine priced second by second on the compute preset selected above, at the rates in the rate card, and a meta-solver is the range between its cheapest candidate and its dearest. Every engine is drawn the same way: the list names them and gives each one's cost at the second under the cursor.Billing, honestly
Is a proven-infeasible result billed?
Yes. The solver worked and the proof is the product: knowing a plan is impossible is exactly as valuable as a plan. It is billed like any solve, for the time consumed.
What if a solve crashes mid-run?
Not your problem, not your bill. The solve is detected and re-queued automatically, and picked up again. Only the successful attempt is metered. Our failures are free.
What does cancelling cost?
The seconds consumed until the cancellation, never more. A problem cancelled while still queued is free, and the whole reservation goes back to your balance.
Which compute preset should I pick?
The cheapest one that finishes your model in time. A larger preset costs more per second, so it pays only when the extra cores or memory shorten the solve by more than the multiplier. Each engine names the presets it runs on, and one of them is its default.
How do I verify what I was charged?
Every solution carries its metering report, every ledger entry links its problem, and the formula above is the whole formula. Your bill is reconstructable from data you hold.
Your first solve is five minutes away
Run a real model against a real solver, in the browser. No account, no credit card.