Grok 4.3 Resets xAI’s API Stack Around One Large Context Model

Release Overview

xAI made a more consequential API move than a casual changelog read would suggest on May 15, 2026. The company repositioned Grok 4.3 as the main text model in its stack, while simultaneously retiring a long list of earlier Grok slugs through the May 15 migration guide. That combination matters because it does not merely add another model card. It changes how developers are expected to build on xAI from this point forward.

The practical story is consolidation. The updated docs describe Grok 4.3 as xAI’s flagship model for general text work, with text and image input support, a 1,000,000 token context window, configurable reasoning, and standard pricing of $1.25 per million input tokens and $2.50 per million output tokens. At the same time, xAI says deprecated text slugs such as `grok-4`, `grok-4-fast`, `grok-4-1-fast`, and even `grok-code-fast-1` will route to Grok 4.3 and bill at Grok 4.3 rates. That is a clean signal that the platform now wants one main cognitive engine rather than a fragmented family of overlapping text endpoints.

What Actually Changed On May 15

The models overview and the dedicated Grok 4.3 page together make the change clear. Grok 4.3 is presented as the primary model for chat, coding, and general-purpose text tasks. xAI lists the model as supporting function calling, structured outputs, and configurable reasoning levels of `none`, `low`, `medium`, and `high`. It also exposes regional availability in `us-east-1` and `eu-west-1`, plus published rate limits of 1,800 requests per minute and 10,000,000 tokens per minute. Those are not marketing adjectives. They are deployment facts that directly influence how production teams size traffic, batching, and cost controls.

The migration angle is just as important as the launch angle. xAI’s docs say that several older models were retired at 12:00 PM PT on May 15, 2026, and that requests to deprecated text slugs will redirect to Grok 4.3. For developers, that means fewer SKU choices but also fewer excuses for model sprawl. A platform team that had separate routing for old Grok text, fast Grok, and code-specific Grok can now simplify around one model and a reasoning knob. That is strategically similar to what other vendors have done when they decide the real product is not a giant dropdown of model names but a default model that can stretch across more workloads.

Why Grok 4.3 Matters Beyond Naming

The strongest part of the release is not the branding. It is the operating profile. A 1 million token context window moves Grok 4.3 out of the standard chatbot lane and deeper into long-context agentic work, repository analysis, document-heavy operations, and multi-step tool use. xAI also states on the model page that Grok 4.3 supports both text and image inputs, which makes it more useful in workflows where a model needs to inspect screenshots, diagrams, tickets, or embedded visual artifacts without swapping to a separate multimodal endpoint.

The configurable reasoning modes are also important because they let teams pick latency and cost tradeoffs per task instead of per model. The docs explicitly support `none`, `low`, `medium`, and `high`, which is effectively a runtime control layer over the same core model. That is a more modern product shape than shipping a dozen adjacent variants that users must benchmark individually. For high-volume support automation or fast coding completions, `none` or `low` may be enough. For long-horizon planning, data analysis, or tool-heavy workflows, `medium` and `high` become the more credible paths.

The Ecosystem Signal Is Bigger Than The Model Card

xAI reinforced the launch with a second signal on the same day. In its Hermes Agent integration announcement, the company said Grok subscribers can use Grok 4.3 directly inside Nous Research’s Hermes Agent, alongside Grok text-to-speech and Grok Imagine. That matters because it shows xAI pushing Grok 4.3 into a broader open-agent ecosystem rather than keeping it framed only as a first-party Grok chat model. The Hermes article describes a persistent agent that can run on a computer, sandbox, or VPS and build long-term memory over time. Putting Grok 4.3 into that context suggests xAI sees the model as an agent substrate, not just a prompt-response endpoint.

That open-agent positioning also helps explain why the model page emphasizes structured outputs and function calling. Those features matter when the model is being asked to operate software, call tools, or pass machine-readable responses into downstream systems. In other words, the release is not just about xAI catching up on context window size. It is about making Grok 4.3 the one model that can sit in the middle of coding agents, assistant workflows, support automations, and long-context analysis without forcing developers to constantly switch SKUs.

What Developers Should Watch Next

There are two caveats worth tracking. First, xAI notes that higher context pricing applies for requests beyond a 200K context window. That means the headline 1M-token window is real, but teams still need to treat long-context usage as a budgeting decision rather than a free capability. Second, the docs say Grok has no realtime knowledge unless developers explicitly enable server-side search tools such as Web Search or X Search. That is operationally healthy because it forces teams to separate raw model reasoning from live information retrieval, but it also means production builders need to wire those tools in deliberately.

As a news story, Grok 4.3 is worth covering because it is a product-architecture shift disguised as a documentation update. xAI is simplifying its text stack, pushing older model names out of circulation, exposing a large context window, and signaling that configurable reasoning is the preferred way to tune work rather than hopping between fragmented model families. That makes Grok 4.3 more than another entry in a leaderboard race. It is the new center of gravity for how xAI wants developers to build.