Demo mode — sample data. Editing is disabled.
Access the real dealRoute each request to the best model for the job with mizan/auto.
Auto-routing savings
What auto-routed calls actually cost, versus what the same traffic would have cost on a few reference models.
$0.03
spent in this period across 7 auto-routed requests
$0.12 saved versus routing everything to GPT-5.6 Sol
If always Claude Sonnet 5
$0.08
If always GPT-5.6 Sol
$0.15
If always Grok 4.6
$0.06
Auto Router
Route to the best model for each request using mizan/auto. When enabled, every call to mizan/auto is classified, ranked against your task's top models, and routed — always billed from your wallet.
Model patterns to filter which models auto-routing can route between. Separate patterns with commas or newlines. Supports wildcards (e.g., anthropic/* matches all Anthropic models). Leave empty to use all supported models.
164 models matched
0 = pure quality (best model regardless of cost), 10 = cheapest model wins. Intermediate values blend quality and cost signals.
Once a model is picked for a conversation, keep using it for up to 5 minutes of inactivity — skips reclassification on every message and keeps that conversation's prompt-cache streak intact, even if a later message would classify differently.
Prevent overrides. Ignore per-request plugins settings from your code and always use these account defaults.
Compare against a specific model
Pick any model to see what your auto-routed traffic would have cost had it all gone there instead.
Select a model above to compare it against your actual auto-routing spend.