xAI Rewrote Its Entire Developer Documentation in One Afternoon

Nobody announced it. There was no launch post, no demo reel, no executive quote in a press release. On the afternoon of April 27, 2026, xAI's developer portal at docs.x.ai simply changed — all of it, at once, with a sitemap lastmod timestamp of 2026-04-27T16:52:16.840Z that applies to 106 pages simultaneously. The quickstart was rewritten. The API reference was restructured. A migration guide appeared. A model comparison table was published. The entire docs site was reorganized around a single interface as the primary path. A company that has spent two years being described as a chatbot with an API decided, in one coordinated deploy, to describe itself differently.

The new story xAI is telling through its documentation is not about Grok's benchmark scores or its voice capabilities or its integration with Tesla. It is about being a platform. And the most telling sentence in the entire refresh is not a description of a feature. It is a URL: /inference/legacy. That path now exists in xAI's REST API reference. The Chat Completions API — the interface every current xAI developer is probably using — has been moved to a URL namespace called "legacy." That is not ambiguous.

What changed and what it reveals

The pages that appeared or materially changed on April 27 tell a coherent story when read together. The new /developers/introduction page did not exist in any meaningful form before — it was a stub or missing entirely. Now it frames xAI as a developer platform from the first sentence. The /developers/quickstart guide now leads with grok-4.20-reasoning as the example model and shows working code in five language-and-SDK combinations: Python via xAI's own SDK, Python via the OpenAI SDK, JavaScript via Vercel AI SDK, JavaScript via OpenAI SDK, and cURL. That quickstart works. It is not aspirational documentation. It is the kind of quickstart that a developer can run in under five minutes.

The REST API reference structure is where the organizational intent becomes visible. There are now separate pages for /inference/chat, /inference/batches, /inference/images, /inference/speech-to-text, /inference/videos, /inference/voice, and /inference/legacy. The legacy namespace is a URL-level signal that the Chat Completions API lives in the part of the docs that is about the past. The current and future is everything else.

New pages in the /tools/ section — /tools/citations, /tools/advanced-usage, /tools/tool-usage-details — suggest that citations and advanced tool configuration are now formally documented capabilities rather than undocumented behaviors that developers had to reverse-engineer from API responses. That matters for the enterprise audience that has been watching xAI from a distance, waiting for documentation solid enough to evaluate against a procurement checklist.

One integration detail worth noting: /integrations/hubspot-mcp-setup appeared in the refresh. HubSpot is the first explicitly named enterprise integration in the new docs structure. That is a meaningful signal. xAI has been positioning Grok primarily as a consumer chatbot or a general developer API. The HubSpot MCP page suggests the company is starting to think about horizontal enterprise workflows — specifically, embedding Grok into CRM-powered business processes. That is a different sales motion than "try Grok in your browser." It is the motion that turns a language model into a business tool.

The reasoning effort clarification that will save someone hours of debugging

The /developers/model-capabilities/text/reasoning page, also updated April 27, contains a detail that trips up production integrations and that nobody had documented clearly before. reasoning_effort — the parameter that controls how much compute the model spends on reasoning — does not work on grok-4.20 or grok-4-1-fast. These models reason automatically; there is no dial to turn. reasoning_effort only works on grok-4.20-multi-agent, and when used there, it does not control thinking effort at all. It controls agent count — specifically, whether the model spawns 4 or 16 reasoning agents. That is a completely different parameter behaving completely differently depending on which model you are using, and it was previously undocumented to the point where developers were probably setting it on grok-4.20 and wondering why their output was not changing.

That kind of documentation gap is the difference between a platform that expects developers to figure things out and a platform that has made the investment to explain itself. The April 27 refresh suggests xAI has started making that investment.

Documentation as a strategic signal

Here is the thing about documentation overhauls that product announcements cannot replicate: they are hard to fake. You cannot ship a polished quickstart that works before the code supports what the quickstart describes. You cannot publish a migration guide that accurately maps parameter names before someone has actually built the replacement API. You cannot put 106 pages of restructured content into a sitemap at a single timestamp unless your engineering team actually updated 106 pages in coordination. Documentation is where the gap between "we say we are a platform" and "we are actually a platform" becomes visible.

The gap that xAI had — and that this docs refresh is attempting to close — was the gap between "we have an API that developers can call" and "we have a developer platform that developers can build on." The difference is documentation, examples, migration paths, honest comparisons, and the kind of quickstart that works the first time. The difference is also URL structure: /inference/legacy versus /inference/chat. That is not a subtle distinction in how xAI is framing its own roadmap.

OpenAI spent years maintaining both its legacy Completion API and its newer Chat Completions API as roughly equivalent paths, letting the market decide which to use. Anthropic never had a legacy API to maintain — the Messages API has been the only path. xAI is doing neither. It is maintaining two parallel paths briefly and then drawing a line through the old one immediately, while the developer base is still relatively small enough to absorb the disruption. That is a different kind of API governance than either competitor, and it only works if the documentation is clear enough that developers understand both the why and the how of migrating.

What this means for teams evaluating xAI

If you have been evaluating xAI from a distance — watching Grok's benchmark numbers, keeping it on a shortlist of model providers — the April 27 documentation refresh is a meaningful data point. A company that rewrites 106 pages of developer documentation in one afternoon is a company that has made a decision about where it is going. That decision is legible in the new docs: the Responses API is the platform, the Chat Completions API is the legacy, Grok's reasoning capabilities are front and center, and enterprise integrations are beginning to be formalized.

The documentation is now credible enough to evaluate seriously. The quickstart works. The migration guide exists and explains the differences honestly. The model comparison table does not hide the tradeoffs. Whether the engineering investment behind it all is durable — whether the Responses API will have the reliability, the SLA, and the support infrastructure that serious production deployments require — is a different question that the documentation alone cannot answer. But the docs refresh removes one major objection that procurement teams and technical leads have been holding: "xAI does not have the documentation to evaluate them properly." That objection is now weaker.

The engineers who noticed the April 27 sitemap change are probably the ones already building on xAI seriously. For the broader developer audience, this documentation repositioning will register retroactively — the next time xAI makes news, the people paying attention will check the docs and find a platform that looks like it knows what it is doing. The question is whether what is behind the documentation will match what the documentation promises.

Sources: xAI Docs Sitemap (106 pages updated 2026-04-27T16:52:16.840Z), xAI Docs — Developer Introduction, xAI Docs — Quickstart, xAI Docs — Reasoning Model Capabilities, xAI Docs — Advanced API Usage