The Grok-2 family of AI models has been released in beta by xAI, the AI firm Elon Musk owns.
On Tuesday, the company launched Grok-2 and Grok-2 Mini, introducing new capabilities and improvements in large language models (LLMs). Both models are now available to Premium and Premium+ users of X (formerly Twitter). Notably, the Grok-2 models include image generation capabilities, though powered by a third-party AI model. The company claims that Grok-2 has outperformed GPT-4 Turbo in benchmarks.
Grok-2 AI Models Released in Beta
xAI announced the launch of the Grok-2 AI models in a blog post, describing Grok-2 as a significant upgrade from the previous Grok-1.5, with advanced chat, coding, and reasoning capabilities. The company also introduced Grok-2 Mini, a smaller, faster model distilled from Grok-2. Both models are accessible to X Premium and Premium+ users, who must update to the latest version of the app to use the new AI features.
According to xAI, an early version of Grok was tested on the Large Model Systems Organization (LMSYS) leaderboard under the name “sus-column-r” and outperformed Claude 2.5 Sonnet and GPT-4 Turbo.
The LMSYS official handle confirmed Grok-2’s high performance, ranking second in coding and mathematics and fourth in handling hard prompts. However, no additional technical details about the models, their database, or architecture have been disclosed.
xAI also revealed a redesigned Grok AI interface with new features, including image generation capabilities. This feature is outsourced to the Flux.1 model created by Black Forest Labs. According to a TechCrunch report, the image generation model lacks safety guardrails and can generate images of politicians.
