Perplexity's tokenmaxxing Model Council gives you multiple bot perspectives

Perplexity wants you to stop asking one chatbot for an opinion and start convening a panel instead. On Tuesday the company rolled its five-month-old Model Council feature into Computer, its browser-based orchestration platform, and in doing so turned what was once a rigid, three-model gimmick into something closer to an actual customizable advisory board.

Here's how it works under the hood, according to Perplexity comms chief Jesse Dwyer: an orchestrator model spins up parallel subagents, each running its own full agent harness on the same prompt, completely independently. A wait barrier holds everything until all the subagents finish their work. Then whichever model the user designated as "chair" reads every response and writes a synthesis, calling out where the models agree and where they split. It's less a chatbot answering a question and more a boardroom simulation, minus the coffee and the passive-aggressive email follow-ups.

When Model Council debuted in February, you had no say in the matter — three preset models, whatever combination happened to be available, end of story. Now users can hand-pick between two and eight models from a roster that spans OpenAI, Anthropic and Google's frontier offerings down to open-weight options like GLM and Kimi. Computer also lets that synthesized output get turned into actual deliverables — reports, board decks, whatever a real business meeting might spit out. Perplexity is pitching this squarely at fuzzy, high-stakes questions: legal judgment calls, financial modeling, growth strategy, engineering tradeoffs — the kind of stuff where there isn't a single correct answer and where seeing several models disagree is arguably more useful than seeing one confidently wrong.

Naturally, none of this is free, and the pricing structure is where things get murky. Model Council access used to require an enterprise plan or the $200-a-month Max tier; now it's open to Pro subscribers paying $20 a month too. But Computer bills on a credit system rather than raw token counts, and Perplexity pegs 100 credits to a dollar. A light task like summarizing a document runs 100 to 350 credits; a heavy one, like assembling a pitch deck, can hit 2,275. Nobody at Perplexity would say how many credits an actual Model Council session burns, because — as Dwyer put it — query complexity matters more than which models you picked. Simple questions might cost less than a penny. Complicated ones, the very kind Model Council is built for, could run considerably higher, and you won't know the damage until after the meeting adjourns.