Models

Each agent runs on one model. Pick it from the model picker in the agent editor, or set it in agent.jsonc. If you don’t pick one, the agent uses Sonnet. Every chat and run of the agent uses that model. You pick a model family, not a version. Each family always runs the provider’s current model, so your agents get new versions without any changes. To see which version a family runs today, hover over it in the model picker.

Available models

Intelligence is Major’s relative rating from 1 to 5. Cost is a relative price band from $ to $$$$$. Actual billing is usage-based.

Model	Provider	Intelligence	Cost	Best for
Opus	Anthropic	5 / 5	`$$$$$`	Deep, multi-step reasoning.
Sonnet	Anthropic	4 / 5	`$$$`	Everyday coding and analysis.
Haiku	Anthropic	1 / 5	`$$`	High-volume, lightweight tasks.
GPT Sol	OpenAI	5 / 5	`$$$`	Complex coding and agentic work.
GPT Luna	OpenAI	3 / 5	`$`	Focused, high-volume tasks.
Gemini Flash	Google	3 / 5	`$$`	Sustained agentic and coding tasks.
Gemini Flash Lite	Google	1 / 5	`$`	Fast, high-throughput execution.
Kimi	Kimi	4 / 5	`$$$$`	Long-horizon agentic coding.
MiniMax	MiniMax	3 / 5	`$`	Long-horizon coding and agentic work.
Grok	xAI	5 / 5	`$$$`	Long-running agents, coding, and knowledge work.

Opus, Sonnet, and Haiku are Recommended by Major.

The Platform Agent works differently: people choose a tier (Smartest, Balanced, Fastest) and admins pick the model behind each tier. See Settings > Models.

Get Started

Learn

Build

Settings

Reference

Available models

​Available models

Available models