Identical prompts, no leakage
Both contenders get the same request and the same instructions. Neither model is told it is being compared, so neither can play to the judge.
The method
Comparison you can actually check
Most AI tools hand you one answer and ask for trust. DyadCouncil runs the same request through two different model families, shows both answers in full, then makes a third model defend a decision between them.
Structured answers
Each model must first break the request down into what it actually requires, then answer, then commit to a bottom line. That makes the two answers comparable rather than stylistically different.
An independent judge
You choose the judge after reading both answers. It receives the original request and both answers, and must return a winner, a confidence level and a fixed four-criteria scorecard.
Reasoning stays visible
Where a model exposes a reasoning summary, it is shown inline so you can audit how the conclusion was reached instead of trusting the output alone.
The lineup
Contender A · GPT-5.6 Terra
OpenAI · Balanced frontier reasoning
openai/gpt-5.6-terraContender B · Gemini 3.6 Flash
Google · Fast multimodal reasoning
google/gemini-3.6-flashJudge option · GPT-5.6 Sol
OpenAI · Deepest analysis, slower
openai/gpt-5.6-solJudge option · GPT-5.5
OpenAI · Rigorous, well calibrated
openai/gpt-5.5Judge option · Gemini 3.1 Pro
Google · Independent second opinion
google/gemini-3.1-pro-previewRun a comparison
Accounts and saved comparison history are coming in a later release.