xAI launches Grok 4.3 model

Large language models are turning into work software. Not just chat windows — actual systems that search, call tools, run code, and chew thr(grok.com)d Grok 4.3. xAI first put the model into beta on April 17, 2026, then moved it into the broader developer stack by early May and(docs.x.ai)uld pick. (grok.com) ### What actually launched? Grok 4.3 is xAI’s new general-purpose flagship for the API(grok.com)f Grok 4, 4.1 Fast, and 4.20 variants with a cleaner default. xAI’s docs now describe Grok 4.3 as the model for “everything else” in chat and coding, while older fast models are already scheduled for retirement on May 15, 2026. (docs.x.ai) ### Why is the April 17 date important? Because this was not a single big-bang reveal. Grok 4.3 showed up first as (grok.com)il 17, with xAI saying it matched the scale of Grok 4.20 but used an improved architecture and a December 2025 knowledge cutoff. The broader push came later, when xAI’s developer docs and pricing pages elevated 4.3 into the main API lineup. (grok.com) ### What’s the headline spec? The big number is the context window — 1 mi(docs.x.ai)e text, code, and prior conversation in view than a standard chatbot session, which matters for long documents, investigations, and multi-step agent workflows. xAI lists pricing at $1.25 per million input tokens and $2.50 per million output tokens. Oracle’s model page mirrors the same 1 million-token limit and notes function calling, structured outputs, and image input support. (docs.x.ai)rtly, but speed is not the whole pitch. xAI calls Grok 4.3 its “most intelligent and fastest” model, but the more meaningful claim is that it is built for agentic reasoning, knowledge work, and tool use. Basically, xAI is saying: don’t think of this as just a better chatbot — think of it as the model you wire into software that has to go do things. (docs.x.ai) ### What does “agentic” mean here? It means the model is supposed to operate with access to tools and a(docs.x.ai) Grok release notes say Grok 4.3 can use the kinds of tools and data a person would want for real work, including a computer where it can write and run code and produce files. xAI’s pricing docs also spell out separate charges for server-side tools like web search, X search, code execution, and file search. (grok.com) ### Why does xAI care a(docs.x.ai)odel companies are no longer just competing on benchmark screenshots — they’re competing on whether developers can build reliable agents for support, finance, research, and internal knowledge work. xAI has been laying that groundwork for months with Grok Business, Enterprise features, and earlier tool-calling pushes like Grok 4.1 Fast plus the Agent Tools API. Grok 4.3 looks like the consolidation step. (x.ai) ### What’s the(grok.com)p input pricing do not automatically mean lower total cost or better results. Tool use adds separate charges, and agent systems can burn tokens fast because they plan, search, inspect files, and loop through tasks. Also, 1 million tokens is actually smaller than the 2 million-token window xAI lists for some older 4.20 and 4.1 Fast variants, so the upgrade is not “bigger in every dimension.” (docs.x.ai) ### Bottom line? Grok 4.3 matters le(x.ai)a product signal. xAI is standardizing around one flagship model for long-context, tool-using work — and trying to make Grok look less like a personality-driven chatbot and more like enterprise infrastructure. (docs.x.ai)