Elon Musk announced on July 8 that xAI will make daily refinements to its Grok Build coding harness and the 1.5-trillion-parameter V9 foundation model, driven entirely by user feedback.

A new 2 trillion parameter model is expected to finish training this month, with availability slated for August 2026. That’s a 33% jump in model size from the current V9 foundation.

What Grok Build actually is, and why the pace matters

Grok Build is a terminal-native coding agent that entered public beta in mid-May 2026. The underlying grok-build-0.1 model became available on May 29. Grok Build operates at over 100 tokens per second, priced at $1 per million input tokens and $2 per million output tokens.

The private beta of Grok 4.5 launched around June 28 at SpaceX and Tesla, built on top of that same 1.5T V9 model but with supplemental training data from Cursor. Preliminary evaluations suggest its performance is competitive with leading models from Anthropic.

The monthly model factory

xAI is targeting the release of entirely new foundation models each month through December 2026. That’s roughly six new foundation models in six months.

Grok 4.5’s use of reinforcement learning to enhance capabilities adds another dimension. The Cursor training data integration suggests xAI is being strategic about where it sources domain-specific expertise.

What this means for investors watching tech and crypto markets

Musk didn’t mention tokens, blockchain, or any digital asset integration in these announcements.

At $1 per million input tokens, xAI is signaling willingness to compete on cost, which could compress margins across the entire AI tooling sector. That has downstream effects on any crypto project attempting to offer AI compute or inference as a tokenized service.

For traders and investors positioned in AI-adjacent crypto tokens, the key metric to watch is whether the 2T parameter model delivers a meaningful performance jump when it becomes available in August.