xAI delays Grok 4.20 training to mid-February

xAI Founder and CEO, Elon Musk, announced that training for Grok 4.20, the next major update to xAI’s flagship AI model, has been delayed to mid-February. The decision comes after power uptime issues at the Colossus facilities, caused by extreme cold weather and construction equipment damaging power lines.

This marks the latest shift in the timeline for Grok 4.20. In November, Musk described it as a “major improvement” and suggested it might arrive by Christmas. He indicated a release in approximately three weeks, which would have placed it in early January 2026. The new mid-February target reflects ongoing challenges in scaling AI training infrastructure.

Grok 4.20 builds on the Grok 4 series, which began with Grok 4.1 in November 2025. That version introduced advanced features like the Agent Tools API, enabling direct access to real-time data, web browsing, and code execution. Grok 4.1 Fast was highlighted for its efficient inference and strong tool-calling performance.

xAI Grok 4.20 training delayed to mid-February due to external issues (Source – X)

Current Grok models have consistently ranked first in independent evaluations, including overall performance on the OpenRouter Leaderboard, programming tasks on Kilo Code and Roo Code, tool calls, creative writing, and factual accuracy. These achievements underscore xAI’s focus on building capable, truth-seeking AI.

Specific benchmarks for Grok 4.20 remain unreleased, as training is still underway. Musk has positioned it as a significant step forward, with Grok 5 planned for later in 2026.

The delay highlights the real-world hurdles of rapidly expanding AI compute resources, even as xAI pushes aggressive development. Users anticipate Grok 4.20 will maintain the model’s reputation for unbiased reasoning and high performance across diverse tasks.