Grok 4.20 Beta 2 fixes hallucinations and follows orders better
xAI just rolled out Grok 4.20 Beta 2, focusing on tighter prompt adherence and fewer capability hallucinations. The update also improves LaTeX scientific text rendering, image search precision, and multi-image generation reliability. While official benchmark numbers haven't dropped yet, it marks another quick iteration to keep Grok on track.
Grok just dropped an update with Grok 4.20 Beta 2, tackling two of the most frustrating traits in modern AI: ignoring user constraints and hallucinating answers.
Polishing the rough edges
According to xAI's brief announcement, the update centers on two main fixes: improved instruction following, so the model actually sticks to user prompts, and reduced capability hallucinations for more factual answers.
Under the hood, xAI also noted better scientific text rendering with cleaner LaTeX output, higher precision when triggering image searches, and improved reliability when rendering multiple images.
Still, xAI shared no benchmark scores, technical details, or broader rollout schedules alongside the post, keeping the model firmly in beta territory.
Why it matters
We think this is a pragmatic, no-nonsense polish rather than a splashy flex. Focusing on prompt compliance and hallucination reduction addresses the everyday friction that keeps users from trusting an AI model.
For researchers and power users who rely on strict formatting or LaTeX math rendering, these targeted tweaks should make daily workflows noticeably smoother.
That said, without published benchmarks, this looks more like an iterative bug-fixing patch than a major leap forward. Grok still needs to prove in real-world hands that it has actually kicked its hallucination habit.
Sources (1)
- x.com(opens in a new tab)Originalx.com