> ## Documentation Index
> Fetch the complete documentation index at: https://docs.ninjachat.ai/llms.txt
> Use this file to discover all available pages before exploring further.

# Ensemble

> Three models answer, a fourth synthesizes. Consensus-driven responses.

Ensemble runs your prompt through 3 models in parallel, then uses a 4th call to synthesize the best answer. Use it when accuracy matters more than speed or cost.

## Request

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

## Response

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

## Variants

| Model ID           | Models used                                | Cost       |
| ------------------ | ------------------------------------------ | ---------- |
| `ensemble`         | GPT-5 + Claude Sonnet 4.6 + Gemini 3.1 Pro | \$0.04/req |
| `ensemble-quality` | GPT-5 + Claude Opus 4.6 + Gemini 3.1 Pro   | \$0.05/req |

## When to use

* High-stakes decisions where hallucinations are costly
* Getting consensus across model families
* Verifying that an answer isn't model-specific bias

Ensemble accepts all standard [chat parameters](/chat) — temperature, max\_tokens, sessions, etc.