Models
Polaris uses Anthropic Claude models through Amazon Bedrock. Requests stay within your organization’s AWS account, and every model call is tagged so usage and costs can be tracked.
Models offered
Section titled “Models offered”| Model | Best for | Price per million tokens (input / output) |
|---|---|---|
| Claude Opus 5.5 | The hardest analysis, long multi-step questions. | $4 / $20 |
| Claude Sonnet 5.5 | Most everyday analysis. A good default. | $2 / $10 |
| Claude Haiku 5.5 | Quick lookups and simple questions. Fastest and cheapest. | $0.10 / $0.50 |
| Claude Sonnet 4.6 | Background work such as chat titles and summarizing long chats. | $3 / $15 |
Prices are the ones Polaris uses to estimate costs. Cache reads and writes have their own prices.
Users pick a model in the chat input, or leave it on Auto to use the default. See Your first chat.
Choose default models
Section titled “Choose default models”Go to Settings › Agent › Models.
- LLM Configuration lists the model provider (Amazon Bedrock) and the models enabled for the project.
- Default models lets you choose One default for everything or A model per task:
- Default chat model: what Auto uses in chats.
- Default for background tasks: titles, memory, summaries.
- Subagent model: the model subagents use. If unset, they use the chat’s model.
The model that answers questions asked through MCP (ask_pda) is set on the Polaris MCP page. See MCP access.
Budgets
Section titled “Budgets”Go to Settings › Budget to cap what the project spends on models. Set a limit per provider and a period (day, week or month), and an overall project limit. When a limit is reached, admins get an email and an in-app notification. See Usage and costs.
Not available
Section titled “Not available”Voice input needs a transcription model, which this deployment does not offer.