Moa — Peer review for every prompt

By default we pick the model best suited for synthesis, drawn from the same frontier models in your plan tier — Pro subscribers get a Pro-tier synthesizer, Max subscribers get a Max-tier one. If you'd rather decide for yourself, Settings lets you choose which model synthesizes the final answer (and moderates in Debate mode). Leave it on the default and you get our recommended pick without having to think about it.

For most people, Pro is all you need — and may even be the better choice. The Pro models are fast, extremely capable, and well suited for research, fact-checking, and everyday analysis. Max is designed for tasks that genuinely require deep reasoning and extended thinking: complex logic problems, multi-step analysis, or high-stakes decisions where you want maximum depth. If you're unsure, start with Pro.

You could — but then you're the synthesizer. You'd need to read three full responses, mentally compare them, spot contradictions, and decide which reasoning to trust. Moa does this automatically with a dedicated aggregation step backed by peer-reviewed research on multi-model synthesis. The result is faster, more consistent, and catches things you'd likely miss when scanning three responses side by side.

All three models run in parallel, so the first phase takes about as long as a single model response. The synthesis pass adds a few extra seconds on top — it's a lightweight aggregation step, not a full research pass. Total response time is roughly comparable to a normal chat response, plus a short wait for the synthesizer to do its work.

Yes. We do not sell your data or use your conversations for advertising, and no conversation data is used for model training — by us or by the AI providers we use. We do store conversations and limited usage and synthesis metadata so we can run, secure, and improve Moa, including giving you history and understanding how the models agreed or differed. You can request full deletion of your account and data at any time. For full details, see our Privacy Policy.

That's when Moa is most valuable. The synthesizer evaluates each position on the strength of its reasoning, not on majority vote alone. If two models agree but the third found something they missed, the stronger argument wins. For factual disputes, the synthesizer can run additional fact-checks via web search. The Venn diagram and color-coded attribution show you exactly where agreement and divergence occurred — and importantly, the models are given anonymous, randomized labels during synthesis, so there's zero bias toward any particular provider.