# LLM Counsel

> Many minds. One considered answer. LLM Counsel is a multi-model AI deliberation platform: query 3-12 frontier LLMs in parallel, have them peer-review each other anonymously, and a designated chairman synthesises one answer with confidence scores and reasoning traces. OpenAI-compatible API.

## How it works

- Stage 1 — Each council member answers the question independently. No model sees the others.
- Stage 2 — Each member anonymously ranks the other answers for accuracy and insight. Labels A/B/C/D prevent reputation effects.
- Stage 3 — A designated chairman (never also a council member) synthesises one answer from the responses and the rankings, with confidence scores and per-model attribution.

## Core pages

- [Homepage](https://llm-counsel.com/): What Counsel is and how it works.
- [Benchmarks](https://llm-counsel.com/benchmarks): Live scores for frontier models across Intelligence Index, GPQA Diamond, MATH-500, LiveCodeBench, MMLU-Pro, AIME 2025 — plus a Council Composite (the max score per suite across a user-selected council).
- [Models](https://llm-counsel.com/models): The frontier-model catalog and pre-built council presets (Reasoning, Code Review, Research, Cost-Aware). Councils hold up to 12 members plus 1 chairman.
- [Confidence & Transparency](https://llm-counsel.com/confidence): How council responses surface agreement %, per-model attribution, and reasoning traces.
- [FAQ](https://llm-counsel.com/faq): Credits, pricing, rollover, and billing questions answered.

## Comparisons

- [vs LLM Routers](https://llm-counsel.com/vs-routers): Routing picks one voice; Counsel convenes the room. They solve different problems.
- [vs Single-Model AI](https://llm-counsel.com/vs-single-model): Why one perspective dressed up as an answer is not counsel.
- [Use Cases](https://llm-counsel.com/use-cases): Legal, financial, medical, and technical decisions where being wrong costs more than asking twice.

## Trust

- [Methodology](https://llm-counsel.com/methodology): The research (MIT 2024 multi-agent debate paper, ensemble-learning parallel), the mechanism (anonymous peer review + designated chairman synthesis), and the Karpathy origin.

## Pricing

Everything runs on credits: $1 = 100 credits, and credits never expire — they roll over month to month on every plan, surviving cancellation and downgrades. New accounts start with 500 free credits. Credits are non-refundable.

- **Pay-As-You-Go** — $0/mo. Pay per query in credits. $8 minimum top-up (= 800 credits).
- **Pro** — $28/mo. 3,360 credits added every month (~17% cheaper per credit than PAYG). Unused credits roll over.
- **Scale** — $280/mo. 38,000 credits added every month (~26% cheaper per credit than PAYG). Unused credits roll over. Priority support.

## Honesty policy

We never fabricate benchmark numbers. Blank cells mean a lab has not published a score and no third-party benchmark has run them yet.

## Build

- [Sign up](https://llm-counsel.com/signup): Create an account.
- [Sign in](https://llm-counsel.com/login): Existing account.