Ask several models. See where they agree.

• 2 to 6 models of your choice

  Pick the models from the catalog or keep the suggested set. The first model you pick writes the synthesis; if that model cannot write it, another model may.

• 6 formats

  General, Research, Coding, Creative, Decision and Debate. The Idea validation template is a Decision with a question skeleton about your idea.

• Independent answers, anonymous critiques

  Each model answers on its own first. Then every model reads the others' answers without knowing whose they are, and critiques them.

• One result

  A synthesized answer, model agreement, areas of disagreement, unique insights and risks, with every model's answer and critique to read.

• Confidence based on model positions

  How many models took the majority position and how close their own ratings are, with the method shown next to the number.

• Files and hand-offs

  Attach up to 10 files, start from a chat with Ask consensus, take a result to Code with Build this, or save it to a project.

1. 01

   Ask the question

   Write what several models should weigh in on, choose a format and 2 to 6 models. The estimated cost shows before you start.

2. 02

   The models answer and critique

   Each model answers on its own, then reads the others' answers without knowing whose they are and critiques them.

3. 03

   Read one result

   One model writes the synthesis: the synthesized answer, model agreement, areas of disagreement, unique insights and risks.

• Claude Opus 5.5
• GPT-6 Sol
• Claude Sonnet 5
• Gemini 3.8 Flash
• GPT-6 Luna
• DeepSeek V4.1 Flash
• DeepSeek V4 Pro
• Claude Opus 5
• Grok 4.7
• Kimi K3
• Qwen3.8 Max
• GPT-6 Astra
• GLM 5.3
• Mistral Medium 3.5
• Claude Fable 5.1
• Gemini 3.6 Flash
• GPT-5.6 Sol
• Qwen3.8 Flash
• Kimi K2.7 Code
• Qwen3.7 Plus
• GPT-5.5
• Mistral Large 3
• Claude Opus 4.7
• Devstral 2
• Gemini 3.5 Flash Lite
• GLM 5
• Gemini 2.5 Pro
• Llama 4 Maverick
• DeepSeek V3
• Qwen-Plus
• Kimi K2
• Mistral Large
• Sonar

Consensus starts with a suggested set of models, and you can change any of them; a model you pick is not replaced, and one that stops answering is left out of the session. Only the synthesis can pass to another model: if the first model you picked cannot write it, ewpire hands it to another model when one is available, and that call is charged as part of the session.

• Claude Opus 5.5: one model call with 2,000 input tokens and 1,000 output tokens costs 0.02954 credits – 0.00844 for the input and 0.0211 for the output.

Counted from the prices in the model catalog, where one credit is $1. The exact charge of each request appears in Usage.

• 2 to 6 models per question.
• A question can be up to 32,000 characters, with up to 10 attached files.
• An attached file can be up to 20 MB, and a model is given up to 200,000 tokens of its text.
• A confidence is calculated for the General, Coding and Decision formats. For Research, Creative and Debate none is shown, because those answers are not meant to agree.
• A workspace can have 50 image, video, music and Consensus jobs queued or running at once.

• It does not tell you what is true. A consensus shows where models agree and disagree, and models can be wrong even when they agree.
• It does not give professional advice. It is a structured comparison of what several models say, and the decision stays yours.
• It does not make up a confidence: when fewer than two models state a position, none is shown.

• Consensus vs a Single Model: What the Research Actually ShowsDoes asking several AI models beat asking one? The measured answer, with sources: where consensus helps, where models agree wrongly, and when one is enough.
• Can ChatGPT Validate Your Startup Idea?One model's yes is tuned to please you. What sycophancy research found, why a single answer proves nothing, and how to make AI attack your idea instead.

• Docs
• FAQ on the modes
• Pricing
• Model catalog
• Idea validation