And then there were 5 - Grok-2 takes a bow

xAI's Grok-2 emerges in beta - how does it compare against its competitors?

Grok-2 – An Overview

xAI, the GenAI company affiliated with Twitter / X, has just released a preview version of its latest foundation model, Grok 2. While currently only available for subscribers to Twitter’s premium platform, its release still signifies that a major new player in the GenAI space has arrived.

According to xAI’ published benchmarks, Grok-2 is showing strong across-the-board performance in general understanding, coding, reasoning, writing and various general various capabilities, on par with GPT-4 and other SOTA models. Grok-2 also incorporates real-time updates from the Twitter / X platform, although that capability is still in active development and results are somewhat inconsistent at this stage.

Grok-2 also features advanced image generation capabilities through incorporation of the FLUX.1 open-source image generating model. This inclusion seems to be generating the most consumer buzz as the generated images are photo-realistic and, for now, seemingly free of Grok-2’s competitors’ copyright guardrails.

Grok-2 comes in two flavours (mini for fast performance and low price, and the state-of-the-art model that we are discussing here). An Enterprise API is also promised, but detailed technical writeup, weights, a production release, and access beyond the X Premium platform are still areas where the company has not yet published its roadmap or intentions.

Market Analysis

Remember when in 2023 we really only had OpenAI as the most advanced LLM? Over the past year we have seen the emergence of five big players in the GenAI ecosystem: OpenAI with ChatGPT, Anthropic with Claude, Meta with Llama, Google with Gemini, and now xAI with Grok-2.

Getting a handle on where each of these earlier competitors stands is straightforward. OpenAI’s early release and technical superiority have helped it lead and maintain its default position as the model to beat. And while its technical superiority if, for now, slowly eroding (with Claude 3 especially giving GPT-4 a good run for its money), OpenAI still remains the gold standard in the industry. Claude is improving by leaps and bounds and is certainly stealing some of OpenAI’s thunder, especially for coding use cases given its excellent Artifacts and Project features. Meta’s Llama 3.1 is also an excellent frontier model with an incredibly appealing open-source license (we discussed this in more detail in a previous post here). And Google continues to invest in Gemini, offering the industry’s largest context window by far and incorporating the model’s capabilities into Android and Google phones.

So where does Grok-2 fit into this landscape? The model is competitive with its peers, especially with its image-generation and real-time tweet integration features. However, uncertainties regarding its availability beyond the X platform, potential ethical concerns surrounding its image model, and the lack of detailed technical specifications give some reason for pause.

Additionally, xAI has indicated that a Grok-3 model is in development, expected to debut next year, potentially timed to compete with upcoming releases like GPT-5. Whether xAI will extend Grok-2's availability beyond the X ecosystem remains to be seen, but doing so would likely enhance its appeal. The current limited access may reduce its attractiveness to a broader audience, despite its advanced capabilities.

However, despite these reservations, the beta release of Grok-2 signals a significant move in the AI industry, establishing a "Big 5" market of generative AI models. The coming months will be crucial in determining how these competitors will evolve, especially as deep-pocketed companies continue to invest heavily in this rapidly growing field.