Grok 4.6
Create an agent or open an existing agent in the builder, then follow the steps below. Model and feature availability may depend on your plan.
What it does
xAI's frontier model joined the selector, with configurable reasoning effort up to xhigh on agentic runtimes.
Before this
An addition to the selector. Grok had been available since 4.3 in May, but 4.6 is the first Grok to expose Maximum reasoning effort, which is what makes it worth the switch on long agentic runs.
Why it matters
It is built for coding, tool calling, and long agentic runs — the cases where a cheaper model stalls halfway through a chain of actions.
The specs
xAI's frontier model, tuned for coding, tool calling, and agentic runs that chain many steps together.
- Context
- 500,000 tokens in and up to 500,000 out — an unusually symmetric budget.
- Input
- Text and images.
- Actions
- Supported.
- Reasoning
- Fast ($$) through Maximum ($$$$$); it is one of only a handful of models that expose the top level.
- Plans
- Every plan.
Maximum effort is worth the credits on an agent that has to stay coherent across a long chain of actions, where a cheaper model tends to lose the thread halfway and start improvising.