Grok

Summary

Organization	xAI
Model name	Grok-0 33B (Aug/2023) Grok-1 314B MoE (Nov/2023) Grok-1.5 (Mar/2024) Grok-2 (Aug/2024) Grok-3 3T MoE (Feb/2025) Grok-4 3T MoE (Jul/2025) Grok-5 6T MoE
Internal/project name	TruthGPT
Model type	Multimodal (text, vision)
Parameter count	See What’s in Grok?
Dataset size (tokens)	See What’s in Grok?
Training time (total)	See working, with sources.
Release date (public)	2023–2025
Playground	https://x.com/i/grok

Grok Updates

15/Nov/2025: xAI CEO to Ron Baron:

It will be the largest model, to the best of our knowledge. So this [Grok-5] is a 6 trillion parameter model, whereas Grok-3 and -4 are based on a 3 trillion parameter model. Moreover, the 6 trillion parameters will have a much higher intelligence density per gigabyte than Grok-4.

I think this is an important metric to think about: intelligence per gigabyte and intelligence per trillion operations. We’ve learned a lot. So the quality of the data that we’re training on with Grok-5 is missionary.

It’s also inherently multimodal, so it’s text, pictures, video, audio. It’s going to be much better at tool use and, in fact, creating tools to be more effective at answering questions and understanding the world. Its vision will be extremely good. It’ll have real-time understanding, which I think is a really fundamentally important thing that other AIs can’t understand—real-time video. I think if you can’t do that, which humans can obviously do, you really can’t achieve AGI. (Interview with Ron Baron, 15/Nov/2025)

Viz. Journey to Grok-5 (2023–2026).
Download PDF

21/Aug/2025: xAI CEO: ‘Grok 5 begins training next month [Sep/2025].’ (Twitter, 21/Aug/2025)

21/Aug/2025: xAI CEO: ‘Wait until you see Grok 5. I think it has a shot at being true AGI. Haven’t felt that about anything before.’ (Twitter, 21/Aug/2025)

9/Jul/2025: Grok-4 launch:
Grok-4 Heavy on HLE text-only subset: 50.7%

Grok 4 Heavy HLE score: 44.4%

Grok 4 Heavy GPQA score: 88.9%

Grok-4 Heavy ALPrompt scores:
Grok-4 Heavy on 2025H1=2/5
Grok-4 Heavy on 2025H2=1/5

Grok-3 launch:
Grok-3 ALPrompt scores (base model/non-reasoning):
Grok-3 on 2024H1=5/5
Grok-3 on 2024H2=2/5 (4/5 with reasoning)
Grok-3 on 2025H1=1/5. First frontier model to consistently admit that it doesn’t know the answer to something.

GPQA Grok-3=75.4%, Grok-3 + reasoning + ITC=84.6%
MMLU-Pro Grok-3=79.9%

13/Feb/2025: xAI CEO: ‘Grok 3 has very powerful reasoning capabilities, so in the tests that we’ve done thus far, Grok 3 is outperforming anything that’s been released, that we’re aware of, so that’s a good sign’ (Reuters, 13/Feb/2025)

9/Feb/2025: On Chatbot Arena (formerly LMSYS), ‘Chocolate’ model (said to be Grok-3) performs very poorly. For ALPrompt2024H2, GPT-4o=4/5. Chocolate=1/5. 

8/Dec/2024: xAI developer Chris Park: ‘Grok 2 + Aurora [xAI’s own image generator] is now available with your X app in the model selector. Oh, by the way, Grok 3 is coming.’

Dataset

Timeline to Grok

Alan D. Thompson is a world expert in artificial intelligence, advising everyone from Apple to the US Government on integrated AI. Throughout Mensa International’s history, both Isaac Asimov and Alan held leadership roles, each exploring the frontier between human and artificial minds. His landmark analysis of post-2020 AI—from his widely-cited Models Table to his regular intelligence briefing The Memo—has shaped how governments and Fortune 500s approach artificial intelligence. With popular tools like the Declaration on AI Consciousness, and the ASI checklist, Alan continues to illuminate humanity’s AI evolution. Technical highlights.