Descriptions:
Wes Roth covers the same-day release of Grok 4.6 from xAI, arguing it marks the first time a Grok model has reached parity with the best offerings from Anthropic and OpenAI at launch. The video walks through the key technical differences between Grok 4.5 and 4.6: a significantly longer supplemental training run built on a 1.5 trillion parameter base model, incorporating curated engineering data sourced in part from Cursor (following xAI’s acquisition), plus an improved optimizer and training recipe.
Roth then demonstrates Grok 4.6’s coding capability live, showing it build a functional Portal 2 clone — complete with portal mechanics, momentum conservation, 3D character models with reflections, and 11 Labs voice effects — in what he describes as essentially a single extended session using Grok Build. He contrasts benchmark skepticism with his genuine surprise at first-use quality.
The second half covers Groqbot, xAI’s new always-on desktop agent platform for Mac, Windows, and Android, which pairs with Grok 4.6’s strong GPT-Val benchmark scores (outperforming rivals on real-world job tasks across engineering, finance, and travel industries). Roth also discusses xAI’s aggressive release cadence — Grok 4.7 expected within weeks, Grok 5 before year-end — and the model’s pricing at $2/million input tokens and $6/million output tokens, roughly half that of comparable frontier models from OpenAI and Anthropic.
📺 Source: Wes Roth · Published August 13, 2026
🏷️ Format: Review
