Ensemble

When accuracy matters more than speed or cost, don’t trust one model — poll three and synthesize:

POST /api/v1/chat

{
  "model": "ensemble",
  "messages": [{"role": "user", "content": "Is microservices worth it for a 3-person startup?"}]
}

Response

{
  "choices": [{"message": {"role": "assistant", "content": "For a 3-person startup..."}}],
  "ensemble": {
    "models": ["gpt-5", "claude-sonnet-4.6", "gemini-3.1-pro"],
    "synthesis_strategy": "consensus"
  },
  "cost": {"this_request": "$0.040"}
}

The panel is task-aware, not fixed. NinjaChat detects the task type (code, math, creative, analysis, factual, or general) from your prompt and picks the 3-model trio best suited to it — e.g. code tasks add DeepSeek V3, math tasks add o3-mini. The example above shows the general panel (GPT-5 + Claude Sonnet 4.6 + Gemini 3.1 Pro), which is also what you’ll see for prompts that don’t match a more specific task type.

Variants

Model ID	Panel	Cost
`ensemble`	GPT-5 + Claude Sonnet 4.6 + Gemini 3.1 Pro (general task)	$0.04/req
`ensemble-quality`	GPT-5 + Claude Opus 4.6 + Gemini 3.1 Pro (general task)	$0.07/req

Use it for high-stakes decisions where hallucinations are costly, or to verify an answer isn’t one model family’s bias. Accepts all standard chat parameters.

Want to see the individual answers instead of one synthesis? That’s Compare.

Get Started

Chat

Superpowers

Media & Search

Reference

Variants

​Variants

Variants