Plurality
A reasoning layer over 30 frontier models

Stop asking one model. Ask a panel.

Plurality sends your question to many models in a structured round, then returns one reconciled answer — with the agreements, the dissents, and the citations laid out next to it.

plurality / new session
⌘K to focus
~12s · est. cost $0.04 Send
Try one →
STEP 01Routing

What happens when you press send.

Your question is parsed, the panel is fanned out, and each model gets the same prompt — in parallel, so none of them can anchor on another.

Your input
"Should we deprecate our REST API in favor of GraphQL?"
Consensus Reasoners
Sent to · 6 models
01
Claude Opus 4.6
Anthropic
02
GPT-5.4
OpenAI
03
GPT-4.1 Mini
OpenAI
04
Gemini 3.1 Pro
Google
05
Gemini 2.0 Flash
Google
06
Nova Premier
Amazon
Independent rounds.
Every model answers before seeing the others. No anchoring.
Equal weight.
No model gets a louder vote unless you tell it to.
Auditable.
Every prompt, response, and revision is logged in the report.
STEP 02Deliberation

Each model answers, with a verdict and a confidence.

What you see below is roughly what shows up on screen during a session — six independent reads on the same question, returned within seconds of each other.

#ModelVendorVerdictReasoning
01Claude Opus 4.6 Anthropicthinking…
02GPT-5.4 OpenAI
03Gemini 3.1 Pro Google
04DeepSeek R1 DeepSeek
05Llama 4 Maverick Meta
06Grok 4 xAI
Live · animated for this page · loops in 6sSkip ahead to the answer ↓
STEP 03The answer

One reconciled report. Agreements, dissents, citations.

Plurality doesn't average the answers — it shows you exactly where the panel agreed, where it split, and what it cited. You get one document you can read in a minute or audit in a day.

report / 0x4f9b · consensus mode · 6 panelists
verified · 74% convergence
Headline
Don't deprecate. Run both, with a hybrid layer.
The panel split 4 against deprecation, 2 for partial / hybrid. No model voted to fully deprecate.
Where the panel agreed4 / 6
+
REST has long-tail consumer cost.
cited by Opus, GPT-5, Gemini, Llama 4
+
Hybrid layering beats full migration.
cited by Opus, GPT-5, Gemini, Llama 4
+
GraphQL is not universally superior.
cited by Opus, GPT-5, Gemini, Llama 4
Where the panel split2 / 6
Timeline (18mo vs 24mo vs indefinite).
DeepSeek R1 vs. Grok 4 — see transcript
Whether to freeze new REST endpoints.
DeepSeek R1 vs. Grok 4 — see transcript
14 citations2.4s total$0.04 spent
view full transcript →
Pick a mode for your question

A different shape of debate for a different shape of question.

The mode is the rules of engagement — how many rounds, what each model is asked to do, how the answer is reconciled. Seven options. One you'll use most.

The play · 01
Quick Take
01Parallel02Synthesize
Qprompt1234Aanswer
One round. Models answer in parallel, then a synthesis pass reconciles them into a single take.
Use this when…ModeRoundsSpeed
Example question
"Should we price by seat, by usage, or with a flat tier?"
Pipeline
parallelsynthesize
What you'd get back
OpusSeat-based. Budgets are how teams buy; predictability wins enterprise.
GPT-5Usage-based. Pain you create is pain you can charge for, and seat caps stop scaling.
GeminiFlat tier with usage above the line. Floor for budgeting, ceiling for upside.
Headline
Three different shapes, no winner. Pick by your buyer: enterprise procurement (seat), self-serve growth (usage), or simplicity at low ACV (flat).
Pick a panel

Five presets. Or compose your own from 30 models.

The panel is who joins the deliberation. A preset is enough for most questions; build your own when you need a specific mix.

All 30 models · grouped by vendor
Anthropic3
OpenAI7
Google4
Meta3
xAI2
Cohere1
Amazon1
Mistral3
DeepSeek3
Alibaba3
Pricing

Plain plans. Cancel any time.

You can do meaningful work on the free tier. Pro is for daily use; Team is for shared workspaces and audited workflows.

Free
$0forever
  • 50 deliberations / month
  • All 7 modes
  • Up to 6 panelists per question
  • Export to Markdown
  • No card required
Start free
Recommended
Pro
$24/ month
  • Unlimited deliberations
  • All panels including custom
  • Up to 30 panelists
  • Export to PDF, Notion, Linear
  • API access
  • Priority routing
Start Pro trial
Team
Custom
  • Everything in Pro
  • Shared workspaces
  • SSO, SCIM, audit log
  • Custom model providers
  • Slack & MCP connectors
  • Annual contract
Talk to us
Five common questions

Things people ask us.

A single model speaks the same way whether it knows or guesses. You can't tell from the prose. With a panel, agreement becomes evidence, and disagreement becomes a useful signal you wouldn't otherwise have.
Free tier · no card · 50 deliberations / month

Bring your hardest question.

Most users get a first deliberation back in under ten seconds. You'll know within one whether this is for you.