What’s in Grok? (2025)

A Comprehensive Analysis of xAI’s Grok models
Alan D. Thompson
LifeArchitect.ai
January 2025
30 pages incl title page, references, appendices.

The report

Coming end Jan/2025…
(Available exclusively to full subscribers of The Memo.)

Abstract
The xAI Grok series of large language models represents one of the most fascinating disruptions in artificial intelligence history. Since launching in March 2023, the series has evolved rapidly from a 33B parameter prototype to planned frontier models measured in trillions of parameters. The Grok series stands alone in its complete absence of publicly released technical documentation. This report presents the first quantitative analysis of xAI’s closely guarded development process, exploring Grok’s architecture, datasets, tokens, parameters, and capabilities across several major model announcements. Using data from the Twitter platform, and training on Colossus—said to be 2024’s largest AI supercomputer—Grok represents an ambitious project in scale and development speed. Building on the acclaimed reports What’s in my AI? (2022), and What’s in GPT-5? (2024), this analysis reveals the details of xAI’s rapid path to superintelligence.

Contents
1. Overview
2. xAI
2.1. Knowledge transfer through staff migration
3. The Grok series of models
3.1. Grok-0
3.2. Grok-1
3.3. Grok-1.5 (+ Grok-1.5V)
3.4. Grok-2 (+ Grok-2-Vision, + Grok-2-mini, + Aurora)
3.5. Grok-3
3.6. Grok-Video
3.7. Grok-4
3.8. Grok-5
4. Dataset: Twitter posts
5. Dataset: Twitter outbound links
6. Dataset: Everything else
7. Hardware: Colossus
8. Model size estimate
9. Grok + Optimus humanoid robots
10. Conclusion
11. Further reading
Appendix A: Datasets Table (Jan/2025 snapshot)

Cover image
Image generated in a few seconds, on 1 January 2025, text prompt by Alan D. Thompson, via Google Imagen 3-002 (mobile portrait 3:4): ‘black on white data pattern, colossus supercomputer, HDR 3d graphic’

Viz and tables

Viz. Where will AGI be born? Dec/2024.

Chart. Employee headcounts at ‘Big 5’ AI labs. 2025.

<See full report for details.>
Table. Timeline of Grok models announcements. Estimated in italics.

Chart. xAI Grok series: MMLU scores (2023–2025).

Image. Grok-2 + Aurora. Prompt by Musk’s four-year-old son, X: ‘Bunnies flying spaceships in Star Wars with a monster truck.’ Jan/2025. LifeArchitect.ai

<See full report for details.>
Table. Cosmos video dataset calculations. Estimates based on NVIDIA Cosmos (20M hours). LifeArchitect.ai

<See full report for details.>
Viz. Journey to Grok-5 (2023–2025).

<See full report for details.>
Table. Twitter text calculations. Tweets after usernames, links, retweets removed.

<See full report for details.>
Table. Twitter outbound links by domain to 2018. Source: GDELT, calcs: Alan.

<See full report for details.>
Viz. Contents of xAI Grok. Jan/2025.

<See full report for details.>
Table. Dataset sizes needed to align with 20:1 data optimization for models.

<See full report for details.>
Image. xAI Colossus supercomputer racks. Memphis, Tennessee, USA. Oct/2024.

<See full report for details.>
Table. Grok training cost and power draw. H100≈$40k (2024), TPD≈1400W/card.LifeArchitect.ai

<See full report for details.>Image. Tesla Optimus Gen 2 humanoid robot. Dec/2023. Source: Tesla.

Viz. Frontier AI models + highlights (Jan/2025). LifeArchitect.ai

Dr Alan D. Thompson is an AI expert and consultant, advising Fortune 500s and governments on post-2020 large language models. His work on artificial intelligence has been featured at NYU, with Microsoft AI and Google AI teams, at the University of Oxford’s 2021 debate on AI Ethics, and in the Leta AI (GPT-3) experiments viewed more than 5 million times. A contributor to the fields of human intelligence and peak performance, he has held positions as chairman for Mensa International, consultant to GE and Warner Bros, and memberships with the IEEE and IET. Technical highlights.