xAI releases Grok 4.6

xAI shipped Grok 4.6 on August 12, 2026, the latest entry in its frontier model line aimed at coding, agentic tasks, and knowledge work. The model is live now on the xAI API with a larger context window and a new tiered pricing structure that splits cost by prompt length.

What's new

According to xAI's own release notes, "Grok 4.6, SpaceXAI's frontier model for coding, agentic tasks, and knowledge work, is now available on the xAI API." The listing spells out the specifics:

• Context window: 500k tokens
• Modality: text and image inputs, text-only output, with no output token limit
• Pricing: $2 / $0.50 / $6 per 1M tokens (input / cached input / output) for prompts under 200k tokens, rising to $4 / $1 / $12 per 1M tokens above that threshold
• Reasoning effort: selectable at low, medium, high (default), or xhigh

The model appeared in xAI's docs alongside a dedicated announcement page, and was picked up by OpenRouter's model registry the same day, putting it in front of third-party routing and benchmarking tools within hours of release.

Context

Grok 4.6 follows Grok 4.5, which xAI had priced flatly at $2 per million input tokens and $6 per million output tokens, without a cached-input discount or a prompt-length-based pricing split. The new tiered structure — cheaper cached-input pricing, and a price step once a prompt crosses 200k tokens — brings Grok's pricing mechanics closer to the tiered schemes OpenAI and Anthropic already use for their frontier models, where long-context requests cost more per token than short ones.

The release lands the same week xAI introduced Grok Bot, a persistent AI teammate for messaging and workflows, suggesting the company is pushing on two fronts at once: a straight model upgrade for developers building on the API, and a packaged agentic product for end users. xAI has moved through the Grok 4.x line quickly in 2026, with each point release focused heavily on coding and agentic performance rather than general chat capability.

Why it matters

The expanded context window and reworked pricing are a direct pitch to teams running long, tool-heavy agent sessions — coding assistants, research agents, and multi-step workflows that chew through hundreds of thousands of tokens per task. A 500k context window with no output cap, paired with a cached-input discount, lowers the cost of the kind of repeated, context-heavy calls agentic coding tools make.

It also sharpens the three-way pricing competition among frontier labs. As OpenAI and Anthropic have each adjusted pricing this year — including Anthropic's decision, announced the day before, to cancel a planned Claude Sonnet 5 price increase — xAI's move to tiered, length-based pricing signals it intends to compete on cost efficiency for long-context workloads specifically, rather than just on raw benchmark scores. For developers choosing among frontier APIs for agentic and coding use cases, Grok 4.6's combination of context size, reasoning-effort controls, and now more granular pricing gives them another lever to optimize against.