🎯Soloone model answers directly — used for simple, low-stakes queries
Solo model
Cheap & fast. Handles lookups, chat, and anything the classifier scores as simple.
⚔️Dueltwo models answer in parallel, then a judge compares their claims
Combatant #1
First of the two independent answerers.
Combatant #2
Second answerer — ideally from a different provider for real diversity.
⚖️ Duel judge
Compares the two answers and flags disagreements. Should differ from the combatants to reduce self-preference bias.
🏛️Council3–5 models deliberate; a judge synthesizes consensus and flags disputes
Council members = Duel combatants #1 + #2 (set above) plus the member below. Add more via the picker on the Orchestrate page.
Member #3
Third default council voice, joining the two duel combatants.
⚖️ Council judge
Synthesizes the consensus answer and marks single-source claims as possible hallucinations.
🧭Routingthe tiny model that scores every Auto-mode query before routing
Classifier
Rates task type, complexity, and stakes. Runs on every Auto query — keep it cheap.
📡Availabilityping every model with a tiny request — free-tier models in particular come and go
Requires your API key. Paid models cost a fraction of a cent per ping; in Free Mode only $0 models are checked.
Prices are per 1M tokens (input/output) from OpenRouter's public list and used for cost estimates.