Vercel AI SDK Adds Support for Grok 4.6 Model IDs and 'xhigh' Reasoning Effort

What happened

Vercel shipped version 3.0.119 of @ai-sdk/xai, the xAI provider package for its AI SDK, on August 12. The release is a patch update with one notable change: it adds model IDs for Grok 4.6 and adds support for an "xhigh" reasoning effort setting when calling xAI models through the SDK.

The change was tracked under commit 451d2c3 with the description: "feat(xai): Add the Grok 4.6 model IDs and support its xhigh reasoning effort." The release itself, tagged @ai-sdk/xai@3.0.119, points to commit 1b0361c on the main branch of Vercel's ai repository on GitHub.

What this actually is

This is not a new model release from xAI. It is an update to Vercel's AI SDK — the TypeScript toolkit developers use to call multiple model providers (OpenAI, Anthropic, xAI, and others) through a unified interface. The @ai-sdk/xai package specifically handles integration with xAI's API.

Adding "model IDs" means developers can now reference Grok 4.6 by name when configuring the SDK's xAI provider, rather than needing to hardcode raw model strings or wait for official typing support. The "xhigh" reasoning effort addition suggests xAI has introduced a new tier in its reasoning-effort parameter — a setting, common across reasoning models, that trades inference latency and cost for deeper chain-of-thought processing. Existing tiers in comparable APIs typically range from "low" to "high"; "xhigh" would sit above the standard "high" setting.

No pricing, context window, or benchmark information was included in this changelog entry. Those details would need to come from xAI directly, not from this SDK update. This release is purely about developer tooling catching up to changes on xAI's side.

What this means

For developers already using the Vercel AI SDK to build on xAI's models, this update means less friction: you can now specify Grok 4.6 and its xhigh reasoning mode without manual workarounds. It's a signal that xAI has made Grok 4.6 and an expanded reasoning-effort option available on its API, since SDK providers typically add model ID support shortly after a provider exposes it.

However, this changelog entry does not confirm official details about Grok 4.6 itself — no benchmark scores, pricing, or context window figures are given here. Anyone evaluating Grok 4.6 should check xAI's own documentation or announcements for verified specifications rather than treating this SDK patch as the model's official release notes. This is infrastructure plumbing, not a product announcement.