Vercel AI SDK Patch Adds Grok 4.7 Model ID Support

Vercel shipped version 5.0.5 of @ai-sdk/xai, the xAI provider package for its AI SDK, on September 21. The release is a patch update containing a single change: support for a new model ID referencing "grok 4.7."

The commit, tagged 7cf7cee, is labeled feat(xai): add grok 4.7 model ID in the changelog. This means developers using the Vercel AI SDK can now pass "grok-4.7" (or the equivalent identifier) as a model string when calling xAI's API through the SDK's unified interface.

What's actually known

This release note comes from Vercel's AI SDK repository, not from xAI directly. It confirms that the SDK's xAI provider now recognizes a model ID for something called "Grok 4.7," but it does not include any details about the model itself — no context window, pricing, benchmark scores, or capability claims are disclosed in this changelog entry. The AI SDK team regularly adds model IDs as soon as providers announce or preview new models, sometimes ahead of, alongside, or shortly after the provider's own public release notes.

No separate announcement from xAI accompanies this change in the source material provided. Whether "Grok 4.7" represents a new checkpoint, an intermediate update to the Grok 4 line, or a preview/beta designation is not specified here.

Package details

@ai-sdk/xai is one of dozens of provider packages maintained under the Vercel AI SDK monorepo, allowing developers to swap between model providers (xAI, OpenAI, Anthropic, and others) using a consistent API surface. Patch releases like 5.0.5 typically bundle small fixes or additive changes — in this case, a single model ID registration — rather than breaking changes to the SDK's interface.

What this means

For developers building on the Vercel AI SDK, this patch means the plumbing is now in place to call a model identified as Grok 4.7 without waiting for a major SDK version bump. It does not, on its own, confirm feature parity, pricing, or availability of that model through xAI's API — those details would need to come from xAI directly. Teams relying on this SDK should treat the model ID as usable but verify actual model behavior, cost, and limits against xAI's own documentation before shipping production integrations. This kind of quiet, patch-level model ID addition is common ahead of or immediately following a provider's own announcement, and it's often the first public signal that a new model version exists at all.