Perplexity expands Model Council to 2–8 AI bots per query

On Tuesday, Perplexity announced that its Model Council feature—originally introduced in February as a three-model advisory system with no user customization—is now available on its Computer platform with significantly expanded flexibility. Users can now assemble a custom council of between two and eight AI models to independently tackle a single ambiguous question. The available models span frontier labs (OpenAI, Anthropic, Google) and open-weight options (GLM, Kimi), and users can adjust analysis depth to suit their needs.

Perplexity's Jesse Dwyer explained the technical architecture: Model Council functions as an orchestrator that spins up parallel subagents, each operating as a full agent harness and analyzing the problem independently. A "wait barrier" holds until all perspectives are complete, then the user's selected synthesizer model acts as "chair," reading all perspective outputs and synthesizing them while explicitly surfacing where models agree and diverge. This consensus-and-conflict reporting is the feature's core value proposition for scenarios with no single correct answer—legal questions, financial queries, corporate decision-making, business growth modeling, and engineering challenges.

Access has broadened from the previous enterprise-only tier (Pro at $34/month per seat, Max at $271, or individual Max at $200/month) to include individual Pro ($20/month) and Max ($200/month) subscribers. However, Model Council operates within Computer's usage-based credit system, where 100 credits = $1. Pricing varies sharply by task complexity: light tasks like summarizing documents cost 100–350 credits, while heavy tasks such as creating work-ready pitch decks can consume as many as 2,275 credits. Pro users receive 4,000 initial bonus credits; Max customers receive 35,000 initial credits plus a renewing 10,000 monthly. Enterprise Pro gets 8,000 bonus credits and 500 monthly per seat; Enterprise Max starts with 45,000 bonus credits and 15,000 per seat monthly. According to Dwyer, query complexity drives token usage more than specific model selection; simple Computer queries can cost less than a penny, while complex tasks scale higher. Perplexity did not disclose a precise credit figure for Model Council sessions, leaving users uncertain about total costs for multi-model advisory queries.

Perplexity's expansion of Model Council reflects a shift in how AI advisory tools handle decision-making under uncertainty. The original February launch confined users to three fixed models selected by Perplexity; the new Computer integration inverts that dynamic by letting users assemble custom boards of two to eight models and control the analysis depth. This directly addresses a use case the company identifies as key: business scenarios where consensus matters more than a single authoritative answer—legal risk assessment, financial modeling, engineering tradeoffs. By explicitly surfacing where models agree and where they diverge, the feature trades the illusion of a single "right answer" for transparency about where experts (modeled as AI systems) actually diverge, which the company argues builds user confidence.

The pricing model, however, introduces a notable friction. Computer credits are abstracted from raw token usage, making the actual cost of a multi-model session opaque—Perplexity acknowledges that "it's not clear how many credits a Model Council session could burn." A "heavy" task like generating a pitch deck can consume 2,275 credits (roughly $22.75 at the published rate), while simple queries may cost less than a penny. The tiered credit allocations (4,000 for Pro, 35,000 for Max monthly) telegraph that frequent multi-model queries could escalate costs quickly, especially for teams relying on Model Council for routine corporate decisions.

No comments yet. Be the first to share your thoughts!