Models and Options
Each AI agent in Simple Agent Designer is configured with a specific AI model. This page helps you understand the available models and how to choose the right one for each agent's task.
How Models Are Assigned
When you create or edit an agent, you specify the model name in the LLM Model field on the Worker (Agent) card. This field is free text — it is not validated against a list, so the name you enter must exactly match what the AI provider expects.
If you leave the LLM Model field blank, the extension falls back to a per-provider default model that is hardcoded into the extension. There is no single shared default. The default for each provider is:
| Provider | Default Model |
|---|---|
| ChatGPT OpenAI | gpt-5-mini |
| Anthropic (Claude) | claude-haiku-4-5 |
| Google Gemini | gemini-2.5-flash |
| xAI Grok | grok-4 |
| Mistral | mistral-large-latest |
| DeepSeek | deepseek-chat |
| Moonshot Kimi | moonshot-v1-8k |
| Alibaba Qwen | qwen-plus |
| Zhipu GLM | glm-4-plus |
| MiniMax | MiniMax-M2 |
| Azure OpenAI | Determined by the deployment in your endpoint URL |
| LM Studio | Whatever model is loaded in LM Studio |
Because the field is free text, the model tables below are advisory — use them as a starting point and consult each provider's current documentation for the exact identifiers they accept.
Available Models by Provider
ChatGPT OpenAI Models
| Model | Strengths | Best For |
|---|---|---|
gpt-5-mini |
Good balance of capability and cost | General-purpose agent tasks (default) |
gpt-5 |
Most capable GPT model, strong reasoning | Complex analysis, multi-step tasks |
gpt-4.1 |
Strong reasoning, large context | Complex analysis, document processing |
gpt-4o |
Multimodal (text + vision) | Image analysis, PDF processing |
gpt-4o-mini |
Smaller multimodal model | Cost-effective vision tasks |
Model availability depends on your OpenAI account tier and region. Check OpenAI Models for the latest list.
Anthropic Claude Models
| Model | Strengths | Best For |
|---|---|---|
claude-opus-4-1 |
Most capable Claude model | Complex reasoning, large documents |
claude-sonnet-4-1 |
Balanced capability and speed | General-purpose tasks |
claude-haiku-4-5 |
Fastest Claude model (default) | Quick categorization, simple analysis |
Check Anthropic Models for the latest model list and pricing.
Google Gemini Models
| Model | Strengths | Best For |
|---|---|---|
gemini-2.5-flash |
Fast, cost-effective, multimodal (default) | General-purpose tasks, vision |
gemini-2.5-pro |
Most capable Gemini model | Complex reasoning, large documents |
Gemini uses its own API format. Check Google AI models for the latest identifiers.
Other OpenAI-Compatible Cloud Providers
These providers use the OpenAI Chat Completions format. Enter the model identifier from the provider's own documentation; the hardcoded default is shown for reference.
| Provider | Default Model | Notes |
|---|---|---|
| xAI Grok | grok-4 |
See x.ai models |
| Mistral | mistral-large-latest |
mistral-small-latest for cheaper tasks; see Mistral models |
| DeepSeek | deepseek-chat |
deepseek-reasoner for reasoning-heavy work |
| Moonshot Kimi | moonshot-v1-8k |
Also moonshot-v1-32k / moonshot-v1-128k for larger context |
| Alibaba Qwen | qwen-plus |
qwen-turbo (cheaper) / qwen-max (most capable) |
| Zhipu GLM | glm-4-plus |
See Zhipu / Z.ai models |
| MiniMax | MiniMax-M2 |
See the MiniMax platform for available models |
Azure OpenAI Models
Azure OpenAI models depend on what you have deployed in your Azure AI Foundry project. Common deployments include:
| Deployment | Base Model | Notes |
|---|---|---|
| Your deployment name | GPT-4 | Must match your Azure deployment name exactly |
| Your deployment name | GPT-4o | Must match your Azure deployment name exactly |
With Azure OpenAI, the model is determined by your deployment — the LLM Model field should match your deployment name.
LM Studio Models
LM Studio supports a wide range of open-source models. Popular choices include:
| Model Family | Example | Strengths |
|---|---|---|
| Llama | llama-3.1-8b | General purpose, good reasoning |
| Mistral | mistral-7b | Fast, efficient for simpler tasks |
| Phi | phi-3-mini | Small, efficient Microsoft model |
| Qwen | qwen2.5-7b | Strong multilingual support |
The model name depends on what you have downloaded and loaded in LM Studio. Check LM Studio's interface for the exact model identifier to use.
Choosing the Right Model
Cost vs. Capability
Larger, more capable models cost more per token. Use this guide to match model size to task complexity:
| Task Complexity | Recommended Tier | Examples |
|---|---|---|
| Simple | Small/Nano/Haiku | Data categorization, yes/no decisions, simple lookups |
| Medium | Mini/Sonnet | Document summarization, basic analysis, structured data extraction |
| Complex | Full/Opus | Multi-step reasoning, financial analysis, code generation, agent cooperation |
Tool Calling Considerations
If your agent uses tools (the Use Tools setting is enabled), choose a model that supports tool/function calling well:
- ChatGPT OpenAI: All GPT-4 and GPT-5 models support tools
- Anthropic Claude: All Claude models support tools
- Google Gemini: Gemini 2.5 models support tools/function calling
- Other OpenAI-compatible clouds (Grok, Mistral, DeepSeek, Kimi, Qwen, GLM, MiniMax): Most current models support tools — confirm with the provider's documentation
- Azure OpenAI: Depends on your deployed model version
- LM Studio: Tool support varies — check your model's documentation
Context Window
Different models have different maximum context sizes. For workloads involving large documents or long conversation histories, prefer models with larger context windows:
| Model | Approximate Context Window |
|---|---|
| GPT-4.1 | 128K tokens |
| GPT-4o | 128K tokens |
| Claude Opus/Sonnet | 200K tokens |
| Claude Haiku | 200K tokens |
Multimodal Capabilities
If your agents need to process images or PDFs (using the file upload feature):
- GPT-4o / GPT-4o-mini: Native vision support
- Claude models: Native vision support for images
- Google Gemini: Native multimodal (image and PDF) support
- LM Studio: Depends on the model (some support vision)
Configuration Examples
High-Quality Analysis Agent
AI Provider: Anthropic
LLM Model: claude-opus-4-1
Use Tools: YesBest for: Complex financial analysis, multi-step reasoning tasks
Cost-Effective Data Processing Agent
AI Provider: ChatGPT OpenAI
LLM Model: gpt-5-mini
Use Tools: YesBest for: Routine data processing, categorization, standard workloads
Privacy-Sensitive Local Agent
AI Provider: LM Studio
LLM Model: llama-3.1-8b
Use Tools: NoBest for: Tasks where data must stay on-premises, no external API calls
Enterprise Azure Agent
AI Provider: Azure OpenAI
LLM Model: your-gpt4-deployment
Use Tools: YesBest for: Enterprise environments with Azure compliance requirements
Tips and Best Practices
- Start with a smaller, cheaper model and upgrade only if the results are not satisfactory
- Different agents in the same workflow can use different models — use a powerful model for the "thinker" and a cheaper model for the "doer"
- Monitor token usage in process activity logs to understand cost per workload
- When using LM Studio, larger models require more RAM and GPU memory
- Test model changes with a sample workload before applying to production workflows
- The model name must be exact — check your provider's documentation for the correct identifier