Try a decision
Pick what you want to decide, give the models some text, and see how each one answers, how sure it is, and what your system would do next.
Advanced: edit the question itself (JSON)
Route a question to the right GPT model
A small decision model reads the prompt and picks the cheapest GPT model that can answer it well. Then we actually call that model. If the router isn't sure enough, it plays safe and uses the strongest model.
Advanced: the GPT models to choose from (JSON: model → when to use it)
The last model in the list is the fallback when the router is unsure.
Triage a support ticket
Paste a customer message and pick a model. In one request it answers six questions: which team should handle it, why they want to return something, what they want to happen, their tone, whether a human should step in now, and how frustrated they are.
Advanced: edit the six questions (JSON)
Benchmark
Run every example that has a known right answer through the models and count how many each one gets right, and how fast.
Traces
Every call to every model is recorded here: what went in, what came out and how long it took. Click a row to see the full details.