Grok, Elon Musk’s AI, becoming interesting for businesses?

Grok, Elon Musk's AI, becoming interesting for businesses?

XAI’s offer appears more and more relevant in an economic alternative to premium publishers in the sector. With two (large) nuance.

He has everything revolutionized the payment sector (Paypal), the automobile (Tesla), space (SpaceX), telecoms (Starlink), probably soon health (Neuralink) and robotics (Tesla still) … Is it already on the artificial intelligence? XAI, the Elon Musk laboratory specializing in generative AI, offers in April 2025 a range of models adapted to many use cases, all with a fairly competitive token price compared to the models of Openai and even Google.

A full model offer (or almost)

XAI offers three types of model: language, reasoning and image generation, all accessible from a conversational interface on the web with Grok.com (with subscription or not). XAI even has its own Deep Research mode to search for advanced information on the web (such as Chatgpt, Gemini and Perplexity). Here are the main models launched by XAI and their respective features in the API:

Model

Input modality

Output modality

Context size (in tokens)

Reasoning

Grok-3-Beta

text

text

131072

Grok-3-Fast-Beta

text

text

131072

Grok-3-mini-beta

text

text

131072

X

GROK-3-Mini-Fast-Beta

text

text

131072

X

Grok-2-Vision-1212

Text / Image

text

8192

Grok-2-Image-1212

text

picture

131072

Grok-2-1212

text

text

131072

On paper, the offer has everything to please. But in reality, many key features are absent from the models offered in the XAI developer API. For example, in theory, Grok-3 is capable of reasoning before responding. However, in the API, only the mini version of Grok-3 is equipped with it. The model being smaller, the capacities are thus reduced. In addition, the multimodal capabilities of Grok-3 (understanding of images and videos) are not available in the API. To analyze visual elements, it is necessary to use Grok-2-Vision-1212, a version already dating back several months. Likewise on the context: Grok-3 is presented with a context window of a million tokens but only arrives with 131,072 tokens in the API. Finally, even more problematic, the most recent xai, Aurora model of image generation,, is still not available in the API. All without talking about the lack of web research support for all models.

The only truly positive point with its Grok-3 series, XAI offers a Fast mode which serves the model from a faster infrastructure for use cases where latency is essential.

Attractive pricing

This is the real strong point of Xai. The prices charged are reasonable or even really advantageous. The classic version of Grok-3 (Grok-3-Beta) is not really interesting (absence of reasoning and multimodality support). On the other hand, Grok-3-Mini offers a minimal cost for reasoning needs, as we explain here. The model is ideal for making a simple and inexpensive agent. However, we can regret the absence of prompt hide functionality which would further reduce the bill.

Model

Input price (million tokens, $)

Price in output (millions of tokens, $)

Grok-3-Beta

3

15

Grok-3-Fast-Beta

5

25

Grok-3-mini-beta

0.3

0.5

GROK-3-Mini-Fast-Beta

0.6

4

Grok-2-Vision-1212

2

10 with text, 2 with an image

Grok-2-Image-1212

0.07 (by image generated)

Grok-2-1212

2

10

XAI gaps to conquer the market

If XAI already offers economically interesting promises on paper on paper, its offer intended for developers is still too incomplete to be recommended. It is therefore impossible to turn entirely to XAI for its generative AI strategy. A situation all the more regrettable when you know that XAI has internally has all the features to serve your chatbot on grok.com. In the field of pure assistance on the other hand (use of Grok as a daily Copilot), the start-up of Elon Musk is doing quite well. With advanced reasoning features (with Grok-3 therefore), advanced multimodal analysis and advanced web search with Deep Research the offer has nothing to envy to Chatgpt. Only its higher price can put off the self -employed: $ 30 per month for Super Grok against 20 for Chatgpt Plus. Business prices are negotiated on a case -by -case basis depending on the company.

Be careful, as we already explained in September 2024, the XAI models have a quasi-total freedom of expression, in accordance with the will of Elon Musk. If in daily practice the total free speech is not problematic, exhibiting xai models directly with customers still seems dangerous (with a chatbot for example). The use of additional safeguards can however drastically limit risks. For example by using an additional LLM to judge the danger of responses (LLM AS A Judge approach). In addition, in agentic mode, the raw use of XAI models does not seem to pose any major problems (the answers not being exposed publicly).

In short, XAI’s young offer is not yet at the level of an Openai or a Google. But this is not surprising: XAI has only two years of existence against ten for Openai or fifteen for Deepmind. With a more complete API (surely in the coming months) and large actors’ customer testimonies using XAI models, the company has everything to explode in the next two years.

Jake Thompson
Jake Thompson
Growing up in Seattle, I've always been intrigued by the ever-evolving digital landscape and its impacts on our world. With a background in computer science and business from MIT, I've spent the last decade working with tech companies and writing about technological advancements. I'm passionate about uncovering how innovation and digitalization are reshaping industries, and I feel privileged to share these insights through MeshedSociety.com.

Leave a Comment