Model Registry

Agent Orchestrator supports multiple AI models across three capability tiers. Assign the right model to each agent based on task complexity.

Available Models

Logical Name Model Tier Best For
haiku Claude Haiku 4.5 Standard Quick tasks, classification, extraction, formatting
nova-lite Nova Lite Standard Cost-effective processing, summarization
nova-pro Nova Pro Standard General-purpose, balanced performance
sonnet Claude Sonnet 4.5 Advanced Research, analysis, writing, tool use — best all-rounder
opus Claude Opus 4.6 Premium Most capable — complex reasoning, strategy, nuanced writing

Amazon Bedrock Access

The models run on your own Amazon Bedrock, in the Region of your stack. The appliance's six default models are Claude Sonnet 4.5, Claude Opus 4.6, Claude Haiku 4.5, Amazon Nova Lite, Amazon Nova Pro and Amazon Titan Text Embeddings V2 (the last one embeds your documents for the RAG Engine; workflows do not use it).

One-time step for Claude: the Claude models need Anthropic's one-time use-case form, plus one invocation of each Claude model, by an AWS account administrator. Until that is done the console's health panel shows the Claude model as denied. Amazon Nova and Titan need no such step. See Amazon Bedrock model access for the steps.

Pricing

Model usage is billed by AWS to your account (Amazon Bedrock) — AppXen adds nothing on top and does not meter or bill for tokens separately. See the AppXen Platform appliance for how billing works.

Using Models in Workflows

Assign models to agents with the Model: line. If omitted, agents default to Sonnet. 

### Fast Classifier
Categorizes input documents by type.
Model: Haiku

### Deep Analyst
Performs thorough analysis requiring strong reasoning.
Model: Opus

Choosing the Right Model

Use Standard (Haiku/Nova) when:

  • Tasks are straightforward (formatting, extraction, classification)
  • Speed matters more than depth
  • You're processing many items in parallel (fan-out loops)
  • Budget is a primary concern

Use Advanced (Sonnet) when:

  • Tasks need research, analysis, or creative writing
  • The agent uses tools (web search, knowledge base, APIs)
  • Quality and nuance matter — this is the best general-purpose choice

Use Premium (Opus) when:

  • Tasks require deep reasoning, strategy, or complex multi-step logic
  • Output quality is critical (executive reports, legal documents)
  • You need the most capable model available for critical steps

Model Version Pinning

By default, a logical name (like sonnet) resolves to the Bedrock model ID your appliance is set to use. To pin a different Bedrock model ID for one workflow, use YAML frontmatter: 

---
models:
  sonnet: us.anthropic.claude-sonnet-4-5-20250929-v1:0
  opus: us.anthropic.claude-opus-4-6-v1
---

# My Workflow
...

Resolution order: workflow frontmatter > the appliance's defaults. A pinned ID must be a model your account can invoke on Amazon Bedrock.

The appliance's defaults are fixed by its settings and shown, read-only, in the console under Settings → Models, together with each model's Bedrock access status. You do not pick models there; a workflow picks them with Model: lines and frontmatter.