Recommended LLMs for the CAST MCP Servers
Which large language models give the best results with the CAST MCP Servers
Overview
Your AI client sends your questions and the CAST tool results to a large language model (LLM) of your choice. This page lists the models that performed best in CAST’s internal testing, last carried out with MCP Server 3.1.0-funcrel in November 2026.
Recommended models
Multiple LLM providers are supported; however, not all models perform equally when it comes to tool invocation, reasoning, and response reliability. During internal testing, the Claude Sonnet 5.5 and Claude Haiku 5.5 models consistently delivered the best results, especially for a wide range of user queries, from simple to highly complex, with strong consistency and minimal hallucinations. Findings are listed below:
| Provider | Recommended model | Notes |
|---|---|---|
| Claude (Anthropic) | Claude Sonnet 5.5/ Sonnet 4.6 | Most reliable for complex queries requiring multiple tool calls and high-accuracy reasoning. Strong natural-language understanding and minimal hallucinations. |
| Claude (Anthropic) | Claude Haiku 5.5/ Haiku 4.5 | Fast, efficient, and cost-effective. Well-suited for simpler queries while still maintaining good accuracy. |
| ChatGPT (OpenAI) | GPT-6 Sol/ GPT-6.1 Terra | Solid overall models for simple to moderately complex user queries. |
| ChatGPT (OpenAI) | GPT-6 Luna | Lower token consumption and suitable primarily for very basic queries. |