Summary
Wes Roth covers the launch of Grok 4.5 from xAI, the first model trained in collaboration with Cursor following xAI’s acquisition of the popular coding agent. The video argues that Grok 4.5 represents a meaningful competitive catch-up, positioning between GPT-5.5 Extra High and Claude Opus on DeepSWE 1.0, with similarly close results on DeepSWE 1.1, TerminalBench, and SWEBench Pro.
Roth evaluates the model hands-on using Grok Build — xAI’s CLI-based coding agent analogous to Claude Code or Codex — building two complex applications: a physics-accurate 3D sailing simulator with wind mechanics, wave dynamics, and AI-voiced narration via ElevenLabs, and an Elder Scrolls-style RPG with voiced NPC dialogue. Both projects demonstrate Grok 4.5’s ability to self-check visual output, implement nuanced physics, and integrate external audio APIs. The sailing game in particular handles capsizing conditions and a multi-life health system after iterative prompting.
The broader framing is strategic: with Cursor’s training data now baked into Grok 4.5, xAI is directly targeting the coding-agent market that Claude Code and OpenAI Codex dominate. Roth contextualizes Grok 4.5 against Fable 5 (which he considers a category apart at an estimated 1.5 trillion parameters) and GPT-5.6, offering a realistic competitive map for developers evaluating frontier coding models.
📺 Source: Wes Roth · Published July 09, 2026
🏷️ Format: Review
