xAI Grok 4.3
xAI positioned Grok 4.3 as its most intelligent and fastest model, leading with large gains on real-world agentic tasks, a 1M-token context window, and aggressively lower pricing (roughly a 37-58% cut versus the prior Grok 4 generation). Third-party coverage also reported native video input as a headline capability. Notably, xAI shipped it quietly: the model appeared in the grok.com selector without a dedicated launch post, blog announcement, or release-notes entry.
The model is real and live: it is listed on xAI's official model docs and served via the xAI API and Amazon Bedrock. Independent benchmarking (Artificial Analysis) confirms the agentic improvement and the price reduction; the much-touted 'native video input' claim, however, appears only in secondary blogs and is not stated in xAI's official docs or in independent analysis.
Grok 4.3 is a genuine, deployed frontier model, but the way it shipped is the story: no official announcement, no blog post, just a quiet appearance in the model picker. The verifiable wins are agentic task performance and a real double-digit price cut, both corroborated independently. The flashier 'native video input' framing is a marketing-grade claim we could not confirm in any official xAI source, so executives should treat it as unverified until xAI documents it. The pattern here — capability shipped faster than documentation — is increasingly common and means buyers must validate claimed features against their own use, not the launch chatter.
A frontier-tier model that is cheaper and more agentic than its predecessor lowers the cost of automating multi-step work, but the silent, under-documented launch is a reminder that vendor claims now outrun vendor documentation.