# Primaris Model Selection

## Overview

Primaris IT has two approved AI models deployed via **Azure AI Foundry**. You must always use one of these two models. No other model endpoints are permitted under any circumstance.

---

## Approved Models

| Model | Endpoint ID |
|-------|------------|
| Claude Sonnet | `eai-dev-claude-sonnet-4-6` |
| Claude Opus | `eai-dev-claude-opus-4-6` |

---

## When to Use Sonnet

Use Sonnet as the **default** for all tasks unless complexity warrants Opus.

- Research and information gathering
- Planning and task breakdown
- Writing and editing documentation
- Building SOPs, guides, and reports
- Creating presentations and slide content
- Quick data analysis and dashboards
- General productivity and communication tasks
- Summarisation and review

**Default to Sonnet when in doubt.**

---

## When to Use Opus

Use Opus when the task requires deeper reasoning or Sonnet has produced insufficient results.

- Complex multi-step software development
- Architecture design and technical decision-making
- Deep code analysis, refactoring, or debugging across large codebases
- Tasks requiring sustained reasoning across many dependencies
- When Sonnet has already attempted the task and the output quality is insufficient

**Only escalate to Opus based on actual task complexity — not as a default.**

---

## Policy

- These are the only two approved models for all AI development activity at Primaris IT
- Never configure or use any other model endpoint, including public Claude.ai, OpenAI, Gemini, Groq, or any other provider
- Both models are deployed within Primaris Azure infrastructure — prompts and context remain within Primaris systems
- If you are unsure whether a model endpoint is approved, stop and check with the IT Department before proceeding