xAI
xAI Grok 4.20 appears on LMArena, launch imminent
LMSYS Chatbot Arena (widely known as LMArena) users spotted two new models: “Slateflow” and “Tidewisp.” Screenshots shared on X by accounts including @AiBattle_ showed these models self-identifying in LMArena responses as variants of unreleased Grok 4.20.
The AI community on the social media site quickly linked these new models to Grok 4.20, the next iteration following Grok 4 and Grok 4.1, released last year. The sightings follow a pattern seen with prior xAI releases, where mysterious or codenamed models appear in public benchmark and testing platforms for blind testing before final launch.
Another Grok model, “Tidewisp”, is also currently being tested pic.twitter.com/PSLPTW39EE
— AiBattle (@AiBattle_) January 14, 2026
Additional reports mentioned a model codenamed “Pearl” performing strongly in the separate Design Arena, further fueling speculation that xAI is evaluating multiple checkpoints of the same underlying system.
These tests highlighted potential strengths in areas such as coding, frontend web design, and complex reasoning. Some users noted polished outputs in tasks like generating full websites, while others described the models as competitive but not as good as leading alternatives.
Just recently, xAI has opened internal beta sign-ups for Grok 4.20, which revealed strong math problem-solving capability.
The new Grok version is expected to improve agentic behaviour, tool use, and code generalization, upgrading from Grok 4.x versions. Amid these revelations, xAI hasn’t confirmed the models’ identity or a release date. Past previews in LMSYS arenas have typically preceded announcements by days or weeks, suggesting a possible launch in the near term.
Grok’s 4.1 Fast and 4.1 Thinking versions are now available across Grok platforms.
