xAI's Grok 4.3 Arrives as a Budget Frontier Model

xAI is competing on price. On June 15, 2026, Grok 4.3 landed on Amazon Bedrock billed as the cheapest frontier reasoning model from a US lab — a pointed move in a market where capability is converging and cost is becoming the deciding factor.

Cheap, capable, and widely available

Putting Grok 4.3 on Bedrock matters as much as the model itself. It drops Grok into the same procurement pipeline enterprises already use for other models, and pairs frontier-grade reasoning with aggressive pricing. As open-weight models keep closing the gap with the frontier, undercutting rivals on price is one of the few durable ways to win share — and it’s the lane xAI is choosing.

xAI also shipped Grok Imagine Video 1.5, an upgrade to its image-to-video generation with better quality and faster output, available through the Imagine API and its consumer apps. Like most modern image and video generators, it’s built on diffusion techniques.

Grok 5 keeps slipping

The bigger story is the model that hasn’t shipped. Grok 5 — once scheduled for January 2026 — is the largest AI model xAI has publicly described: a roughly 6-trillion-parameter system trained on its Colossus 2 supercluster. Elon Musk has gone as far as claiming a “10% probability” that it achieves the first true AGI, a characteristically bold framing.

The reality has been more sober. Grok 5 remained unreleased through the first half of 2026, and prediction markets put only about a one-in-three chance on it shipping by the end of June. Training and stabilizing a model that large is hard, and slipping timelines are now the norm across every frontier lab, not just xAI.

Where xAI fits

The competitive picture is crowded. Google’s Gemini 3 is pushing reasoning and multimodality, Anthropic’s Claude is targeting long-horizon agents, and OpenAI keeps iterating quickly. xAI’s wedge is a combination of low price, fast iteration on point releases, and tight integration with X’s real-time data and its own massive compute.

The takeaway

Grok 4.3 shows xAI’s near-term strategy clearly: ship capable models, distribute them where enterprises already buy, and win on cost. Grok 5 is the long bet — enormous, ambitious, and running late. For now, the cheapest-frontier-model pitch is the more interesting one, because it targets exactly the pressure every large language model provider is feeling as raw capability stops being a moat.