Feature
Orchestration labs: compare, judge, consensus, escalation
The ClawAI labs that put several models on one prompt — Compare with Judge and Critic, Consensus, Escalation, Best-of-N, Verify, Repair, Pipelines and more.
All features · Last reviewed:
Compare, with a judge and a critic
Compare sends one prompt to several models and shows their answers side by side with latency and token counts. Switch on Judge and an independent model scores each answer against explicit criteria; switch on Critic and it writes down what is weak in each. Compare also runs inside an ordinary thread, so you can check a single answer without leaving the conversation.
Consensus and escalation
Consensus asks two to five models the same question and synthesises one answer from where they agree, flagging where they do not. Escalation starts with an inexpensive model and moves up the chain only when the answer falls short, so you pay for a strong model when the question actually needs one.
Verify, repair, and the rest of the bench
Best-of-N generates several candidates and keeps the strongest. Verify has a second model check an answer for correctness. Repair fixes a specific defect in an existing answer instead of regenerating it. Decompose splits a large task into steps. Role packs hand a problem between role-specialised models, Cost ensemble balances quality against spend, and Pipelines chain several stages into one named, re-runnable workflow.
Each lab is enabled per plan by the operator, and lab runs are metered separately from ordinary chat — Compare, Judge and Critic on their own surfaces, the other labs on the orchestration surface.
Questions people ask
- What is the difference between Compare and Consensus?
- Compare shows every model’s answer side by side and leaves the verdict to you, optionally with a Judge score. Consensus merges the answers into one and flags the points where the models disagree.
- How does escalation save money?
- It starts with a cheaper model and only moves to a stronger one when the answer does not meet the bar, so easy questions never pay for the most expensive model.
- Are the labs on every plan?
- Each lab is switched on per plan by the operator, so availability depends on your plan. The pricing page lists what each plan includes.
Try it rather than take our word for it
Compare, Consensus, Escalation, Repair, Decompose, Best-of-N, Verify, Pipeline, Cost ensemble and Role pack are each shipped with their own page, endpoint and result card.