Last updated October 2026
We wrote ten optimisation problems with uncertain inputs and whole-number decisions. Most have a probability limit the answer must respect, such as at most a 5% chance of an oxygen crash. On an ordinary two-core PC, the ModelRisk Optimizer solved each one in 7 to 56 seconds, and all ten in under 3 minutes.
For seven of the problems we know the best possible plan, because we checked every combination. ModelRisk found it exactly on four, and on two more its plan was within 1.5% and 8.3% of it. On the seventh, and on one problem without a known optimum, its plan went slightly over a probability limit.
All ten problems are free to download in ModelRisk, @RISK and Crystal Ball versions, so you can run them in the optimiser you use today and compare.
Each problem has uncertain inputs and whole-number decisions. Switch between how long each took and how good the answer was.
Time from starting the optimisation to getting the answer back. Every plan tried is a full simulation of the model, 1,000 to 4,000 samples. One PC: Windows 11, 2 processor cores, 8 GB of memory, Excel 2019.
Optimum: the best possible plan, found by checking every combination. % from optimum: how much worse the plan was. Over its limit: the plan breaks a hard limit, such as a 5% risk cap, when checked with a larger simulation.
Each plan scored exactly where the model allows, otherwise from a 20,000-sample simulation.
Run the same ten problems yourself, in ModelRisk and in the optimiser you use today.
Every trial in an optimisation is a full simulation of the model, so the simulation engine sets the pace. On the museum timetable, ModelRisk tried 2,192 plans in 56 seconds, each one a 1,000-sample simulation. Across the ten problems it ran 24 to 83 of these simulations a second.
ModelRisk ran all ten problems on its Turbo engine. Turbo analyses the workbook once, compiles only the cells that change between iterations and runs them on every processor core, without Excel recalculating. Across our wider test library, its simulation phase runs a median of about 200x faster than ModelRisk’s classic engine.*
The search method matters too. ModelRisk’s Auto setting picks an engine from the structure of the model. For these ten whole-number problems it picked the Vertex Detector.
* Simulation phase only: 89 models from our internal test library, same workbook, 1,000 iterations, Turbo against ModelRisk’s classic engine. Excludes the one-time model analysis and opening the workbook. The gain depends on the model.
Try the Optimizer on your own model with the free 15-day trial. The Optimizer and Turbo are included in every ModelRisk licence.
Download the ten test models (0.7 MB zip). Each problem comes in three versions with the optimisation already set up: ModelRisk, @RISK with RISKOptimizer, and Crystal Ball with OptQuest. A scorecard lists ModelRisk’s result for every problem, and the best possible answer where we know it.
Your times will differ from ours on another PC or with other versions, so run both tools on the same machine. The settings in each workbook use the same simulation size, the same maximum number of trials and the same stopping rule, so the problems are set up alike in all three tools.
On two problems its plan went slightly over a probability limit. In snowmaking, the plan gave a 10.8% chance that fewer than five of the six slopes open, against a limit of 10%. In aquaculture feeding, it went over the 5% limit on an oxygen crash.
This can happen with any optimiser that works from simulation. It estimates the probability from its own samples, and a plan close to the limit can land on either side of it.
So check a plan that sits close to a probability limit with a larger simulation before you act on it, or set the limit a little tighter than the real one.
The museum timetable is the largest problem, with 4.4 trillion possible timetables. There ModelRisk’s plan was 8.3% off the best possible spread of visitors.
They come from the example library we wrote for the ModelRisk Optimizer. Each one fits on one screen, and the @RISK and Crystal Ball versions use the same formulas, inputs and limits.
€1,550 at most, per user per year
Optimizer and Turbo included in every licence. Price list
€3,500 per year
The only @RISK edition with RISKOptimizer. Professional, €2,900, doesn’t include it.
Not published
Optimisation comes in the Decision Optimizer edition. Oracle quotes prices on request.
So in ModelRisk the optimiser costs nothing extra. In @RISK it means buying the Industrial edition, €600 a year more than Professional.
Yes. ModelRisk converts @RISK and Crystal Ball models free of charge, and the converter brings RISKOptimizer models across as well. Converted models remain ordinary Excel workbooks that colleagues can open and audit.
Read more about switching from @RISK or how ModelRisk compares with Crystal Ball, and what the ModelRisk Optimizer can do. Then start the free 15-day trial and run one of your own models.
On our ten test problems it took 7 to 56 seconds each, and under 3 minutes for all ten, on a two-core PC. Every plan it tries is a full simulation of 1,000 to 4,000 samples. Larger models take longer.
Yes. Download the ten test models: each comes in ModelRisk, @RISK and Crystal Ball versions with the optimisation set up and the same settings saved. Run them on one PC, then compare the answers with each other and with the known optima in the scorecard.
Yes. The Optimizer and the Turbo engine come with every ModelRisk licence, at most €1,550 per user per year, and the 15-day trial is fully functional. In @RISK, RISKOptimizer needs the Industrial edition.
An optimiser estimates a probability limit from its own simulation, so a plan close to the limit can land on either side of it. It happened on two of our ten problems. Check a plan near a limit with a larger simulation before you act on it, or set the limit a little tighter.
@RISK and RISKOptimizer are trademarks of Lumivero. Crystal Ball is a trademark of Oracle Corporation. OptQuest is a trademark of OptTek Systems, Inc. Vose Software is not affiliated with any of them. Product details from Lumivero’s web shop (10 October 2026), oracle.com and Capterra (August 2026) may change.
Run the test models, or your own, through the Optimizer. The 15-day trial is fully functional, and the free converter brings @RISK and Crystal Ball models across.