Select and Switch Between AI Models
Learning Objectives
After completing this unit, you’ll be able to:
- Identify which AI models are available in Agentforce Vibes.
- Describe the strengths of each type of AI model.
- Apply selection criteria to choose the right model for different development tasks.
- Switch between models using the Agentforce Vibes UI.
Not All AI Models Are Created Equal
An AI model is essentially the “brain” that processes your text and generates code. In a provider-agnostic, multi-model world, it’s critical to look past brand loyalty and match the right model to the right job.
- The challenge: Your development team might be using a powerhouse AI by default for everything, but relying on a single option across the board creates a massive bottleneck.
- Why model selection matters: There are huge performance, cost, and capability differences between models. While a heavyweight model is necessary for complex architectural planning or using it for simple code formatting wastes a ton of credits on tasks that a lighter, faster model could do for a fraction of the price.
Match Models to Tasks
When you set up Agentforce Vibes, you aren’t locked into just one AI model. The platform supports a variety of top-tier LLMs and continuously adds new ones to the lineup.
Available Models
The models available to your team depend entirely on the type of billing SKU you choose.
- Metered: Gives you access to all available models based on the consumption cost (Flex credits).
- Unmetered: Gives you access to included and premium models.
Here’s a breakdown of the model types that are available.
- Included models: These are included out of the box in an unmetered plan and only subject to contractual fair use, meaning developers can use them heavily for everyday coding tasks.
- Premium models: These are higher-tier, more capable models available in unmetered mode. Because they are premium, they do come with a cap: your org gets up to 2,500 requests or 36 million tokens, whichever milestone you hit first.
- High-cost models: These models are built for deeply complex architectural problems and logic. These models use advanced prompts that consume high amounts of data and credits. These models consume high Flex credits.
For more information about AI models supported by Agentforce Vibes, see Large Language Model Support.
Model Versus Tasks
Because the platform operates in a multi-model world, true efficiency comes from matching the exact model strength to the task at hand. Let’s look at use cases to see where each model thrives.
Code-generation tasks:
- The best fit: Lighter, highly optimized developer models.
- Why they work best: They have low latency and can handle framework rules. They are the perfect choice for writing clean Apex logic, generating Lightning Web Components (LWC) scaffolding, or autocompleting boilerplate code in real time without lagging the developer’s integrated development environment (IDE).
Context-heavy tasks such as reading and analyzing an error log:
- The best fit: Models with massive context processing capabilities
- Why they work best: These models can ingest entire repository structures or giant server error logs at once, allowing them to pinpoint hidden concurrency bugs across your entire Salesforce stack without missing a beat.
Conversational tasks such as planning and ideation:
- The best fit: Heavyweight reasoning models
- Why they work best: These models use advanced multi step reasoning. If you ask them to plan a complex system migration, they think sequentially, mapping out metadata changes, reviewing architectural dependencies, and building step-by-step technical blueprints before writing a single block of code.
As you can see, optimizing your resources isn’t about picking the cheapest option—it’s about balancing capability with consumption.
Switching Models
You can change an AI model for an individual chat session or change your default model. What models you can select depends on the financial plan you choose.
Changing Models for a Session
To switch up the AI model you’re using, select the model name in the bottom-left corner of your agent-chat window on your IDE. The dropdown lists all the available models grouped by their vendor. Select a model name from the dropdown menu, and you’re good to go.
The best part is that your choice persists for the rest of your session until you decide to change it again. Plus, you can easily swap models right in the middle of a conversation without losing your place or confusing the AI—it keeps all your chat context perfectly intact.
Best Practices
To protect your budget from unexpected credit spikes and to keep workflows efficient across a large engineering footprint, admins can implement structured governance and regular usage auditing.
Restrict model options for governance: To maintain strict governance, protect budget overruns on metered SKUs, or align with compliance laws, admins can intentionally restrict LLM options.
Scale to a large team: When scaling Agentforce Vibes to a large enterprise engineering footprint, use these operational guardrails.
- Standardize your organizational defaults around lighter, unmetered plans for day-to-day coding. Train your team to reserve high-tier, metered plans strictly as an escalation tool for complex architectural planning and massive debugging tasks.
- Monitor AI-usage metrics by going to Setup | Agentforce Vibes Extension | Usage & Adoption. You can download a running 30-day usage data as a CSV file.