Compare
Simo vs reasoning models
A reasoning model writes a few thousand words before each answer. Simo answers directly. They are not rivals: one is built to decide, the other to think.
Side by side
| Simo | Reasoning model | |
|---|---|---|
| Output | Calibrated probabilities and typed values | Prose you must parse |
| Before the answer | Nothing written | A few thousand words |
| Time per decision | Milliseconds (64 ms for one question on Simo-1) | Seconds |
| Confidence | A calibrated number | “I’m confident” |
| Many questions, one input | Read once; each question is a cheap branch | Each question is a fresh generation |
| Best at | Judgment calls inside a loop | Open-ended, multi-step problems |
How they score
On hard decisions, Simo-1 Pro scores 94.71 on DecideBench hard, next to 91.7 for a leading reasoning model on DecisionBench hard. The benchmarks are separate, and the reasoning model was run in thinking mode. Details on the accuracy page.
Use them together
Put Simo in front. It handles the routine on its own, and passes only the ambiguous cases to the reasoning model: act, think or ask. The slow model is called only when it earns its cost.
Frequently asked questions
Is Simo better than a reasoning model?
For in-loop judgment calls, Simo is faster and returns a calibrated number. For open-ended problems, a reasoning model is the right tool. The best setup uses both.
Why not just use a reasoning model for decisions?
It writes a few thousand words per answer, which takes seconds and costs more, and it gives you prose rather than a probability you can threshold.