April 19, 2026 — xAI announced the beta launch of Grok 4.3, marking a significant expansion of its flagship large language model with 1 trillion parameters, doubling the capacity of its predecessor Grok 4.20. The new model is now available to SuperGrok Heavy subscribers on an exclusive basis, with broader public access anticipated within weeks. According to xAI's official blog, the expansion represents a major milestone in the company's efforts to deliver increasingly sophisticated AI capabilities.
Scale and architecture
At 1 trillion parameters, Grok 4.3 is precisely twice as large as Grok 4.20, enabling the model to process and generate more nuanced responses across a wider range of tasks. This substantial increase in model capacity is expected to improve reasoning depth, contextual understanding, and the handling of complex multi-step problems that challenge smaller language models.
Performance improvements
Early benchmark results reveal substantial gains, particularly in reasoning and coding tasks. Grok 4.3 demonstrates enhanced capabilities in mathematical problem-solving, logical inference, and software development scenarios — areas where scale improvements in large language models consistently yield measurable advances.
Availability and access
Currently restricted to SuperGrok Heavy subscribers, xAI's premium tier, Grok 4.3 beta is expected to reach general availability within weeks. The accelerated timeline reflects xAI's confidence in the model's stability and performance readiness for mainstream deployment.
Why it matters
The release of Grok 4.3 intensifies competition in the advanced AI market, where scale and performance benchmarks have become key differentiators. With a trillion-parameter model, xAI positions itself alongside other major players developing frontier systems, underscoring the industry-wide trend toward larger, more capable AI for demanding applications in research, software engineering, and enterprise decision-making.