Swantide lets you choose which AI model powers your AI Assistant conversations. The right choice usually comes down to one question: is your question broad or specific?
Broad questions (spanning many components) benefit from a stronger model that brings more reasoning depth to synthesizing across your org.
Specific questions (a single component or small set) are well served by a fast, lightweight model that answers quickly.
Most of the time, just leave it on Auto.
The AI Assistant is agentic: it queries your org's metadata with tools on demand (searching, counting, reading components, tracing references) rather than pre-loading a fixed handful of components into a context window. Every model retrieves the metadata it needs as it works. Models differ in speed and reasoning depth, not in a fixed number of components they're allowed to see.
Available models
The model picker offers Auto plus several models from each major provider. Because the exact versions change over time as we benchmark new releases, think in terms of families:
Model | Provider | Best for |
Auto | Swantide-selected | The default and baseline. Best for most specific, focused questions. Swantide picks a fast, capable model based on ongoing benchmarking, and the underlying model changes over time as we test new ones. |
Sonnet | Anthropic | Fast and focused. Specific, single-component questions where speed matters. |
Opus | Anthropic | Deeper reasoning. Broad, multi-component questions and synthesis (e.g. a risk summary spanning several flows). Newer versions are available, so try one when an answer feels incomplete. |
Gemini | Alternative provider. Try when the Anthropic models aren't producing what you need. | |
GPT | OpenAI | Alternative provider. Try when the Anthropic models aren't producing what you need. |
At the time of writing the picker lists Auto plus several specific versions across these families (Sonnet, Opus, Gemini, and GPT). The available versions shift as providers ship new ones, so pick by family and recency rather than memorizing version numbers.
Bigger isn't always better. Stronger models are slower and can over-produce, sometimes generating more documentation than you need. That's another reason Auto is a good default.
When to use each
Use Auto when
You don't have a strong reason to pick a specific model. Auto handles most questions well.
Speed matters more than exhaustive detail.
You want Swantide to pick the right model for you based on our benchmarking.
Use a stronger model (Opus) when
You're asking a broad question that touches many components.
You're synthesizing dependencies across flows, Apex, validation rules, etc.
You're generating comprehensive documentation for a complex process.
Auto or Sonnet gave you an incomplete or shallow answer.
"Explain the entire sales process in this org, including all required fields and validation rules across Opportunity, OpportunityLineItem, and Quote."
For a guaranteed-complete field or validation-rule inventory, the Data Dictionary is your source of truth, built to return every item, and that's the right tool regardless of model. Use the stronger model to analyze the process, and the Data Dictionary to confirm the complete list.
Use a fast model (Auto or Sonnet) when
Your question is about a single component or small set.
You want a fast answer.
You're generating a focused artifact (a design doc, a test script, a code review).
"Write a design document for the flow Create_Renewal_Opp."
Use Gemini or GPT when
The Anthropic models (Sonnet / Opus) aren't producing what you need.
You want to compare outputs across providers for a high-stakes design or audit.
A specific use case responds better to a different model, because model strengths shift as providers release new versions.
How to switch models
Open the AI Assistant in your Swantide org.
Click Change model near the chat input to open the model picker.
Pick your model before sending the prompt.
You can switch models mid-conversation, and the chat history is preserved.
Examples by question type
Prompt | Recommended | Why |
"What does the field | Auto | Single component, simple lookup |
"Write a UAT test script for the flow | Auto or Sonnet | Focused, single-flow scope |
"Write a design document for the flow | Auto or Sonnet | Specific artifact, single component |
"Explain the entire sales process in this org, including all required fields and validation rules." | Opus | Broad, multi-object analysis |
"Auto's answer on this audit felt incomplete. Retry with deeper reasoning." | Opus | Escalation when faster models fall short |
For broad "find every component that references X" tasks, the right answer isn't a bigger model, it's the References tab on that component. See How to Use the References Table to Understand Metadata Dependencies. For enumeration like "list every custom field on Account," use the Data Dictionary with filters. The AI Assistant is your tool for analysis, design, and explanation, and the Data Dictionary is your source of truth for complete, org-wide enumeration.
Tips
Default to Auto. Swantide chose it as the baseline based on our benchmarking. Reach for a specific model only when Auto's answer falls short or your question is clearly broad and deep.
Be specific. A focused question about one flow, profile, or field is where the Assistant is strongest. A whole-org overview question is harder regardless of model, and being specific points the Assistant at the right place faster.
Match reasoning depth to question scope. Stronger models reason more deeply across many components; faster models are quicker and more focused.
Escalate from faster to stronger, not the other way. Start specific and lightweight; switch to Opus or another provider only when the answer is shallow.
Try a different provider for a second opinion. If one family is not giving you what you need on a particular question, try another. Different families phrase and structure answers differently, and a second opinion is cheap.