JEVArena
A visual decision-routing simulator for seeing how a routing layer can classify a task, choose a model tier, and trade cost/runtime against direct fixed-model execution.
difficulty: hard
route: premium tier
decision: 0.19s
direct baseline
↳ mismatch loops
↳ extra tokens
↳ extra runtime
Why it exists
Routing is hard to explain with a single cost number. JEVArena turns the decision path into something visible: prompt difficulty, selected tier, decision time, route distribution, token/cost assumptions and comparison against a direct model.
It is deliberately a simulator, not a claim of live provider measurements.
Current simulator model
Run one prompt or randomized benchmark suites from a 1,500-prompt pool.
See route target, decision timing, total decision cost and live route distribution separately from final model execution.
If a lightweight fixed model is asked to handle a harder task, the simulator increases effective tokens and runtime to represent extra loops, retries and repeated context processing.
The normal web simulator runs without an account or API key, and a portable Windows package is available for offline-style local use.
Try the simulator in the browser.
The public deployment lives directly under pouramin.dev.