Choosing the Right Model for Local Development

Running AI locally can be rewarding—you have full control, no cloud costs, and the flexibility to experiment. But when it comes to agent mode in Explyt, the model you choose will make or break your results.

Small models under 40B parameters may be lightweight and easy to run, but they often fail to deliver in reasoning-heavy workflows.

Why Small Models Fall Short

Agent mode is more than just generating text—it requires multi-step reasoning, tool selection, and structured execution. Models like:

• deepseek-coder-6.7b
• starcoder2-7b
• Qwen3-Coder-30B-A3B-Instruct

may work fine in casual chat tools like LM Studio, but in agent mode they struggle with complex, tool-driven tasks.
Read more: How Agent Mode Boosts Your Workflow

Recommended Models

If you need strong local performance in agent mode, go for:

• 30B+ parameter models
• Tool use support enabled

Example: qwen3-235b-a22-2507-instruct — large-scale, reliable, and optimized for multi-step tasks.
Learn more about tool use: Tool Use Guide for Explyt

The Easy Route: Explyt Built-In Provider

To skip the trial-and-error process, you can use the built-in Explyt provider. We’ve already selected a proven model that works across all features. Benefits include:

• No setup—works instantly
• Free trial before committing
• Guaranteed compatibility with all Explyt tools

➡ Using Cloud Models in Explyt

When Stuck With Small Models

If hardware limits you to smaller LLMs, we recommend:

• Disabling tools in agent settings and using chat mode
• Leveraging features that work well with small models:
  • Test generation from requirements
  • Test generation from execution
  • Basic chat features

---