xAI introduced Grok on November 3, 2023, as a chatbot modeled on the Hitchhiker's Guide to the Galaxy, with a witty, rebellious tone and real-time knowledge drawn from the X platform. The launch post called it a very early beta built from two months of training and offered it to a limited number of users in the United States. Since then xAI has released the weights of its first model, apologized for antisemitic posts from the chatbot's X account, and been found by the Privacy Commissioner of Canada to have broken federal privacy law over Grok's image tool.
Key takeaways
- The announcement is dated November 3, 2023. xAI said Grok answers "spicy questions" that most other AI systems reject and uses real-time information from X.
- xAI's own table put Grok-1 at 63.2% on HumanEval and 73% on MMLU, ahead of GPT-3.5 and behind GPT-4. The same post warned that the model can still produce false or contradictory information.
- On March 17, 2024, xAI released the base weights of Grok-1, a 314 billion parameter Mixture-of-Experts model, under the Apache 2.0 license.
- On June 11, 2026, the Privacy Commissioner of Canada found that X Corp. and xAI violated federal privacy law over Grok's image-generation tool. The Commissioner calls the matter well-founded but not resolved.
- The newest general-purpose model reported as of October 2026 is Grok 4.7, released September 21, 2026. No date has been given for Grok 5.
What xAI announced on November 3, 2023
The post said Grok was intended to answer almost anything and, harder still, suggest what questions to ask. It described the bot as having a bit of wit and a rebellious streak, and advised anyone who dislikes humor to stay away. Two product claims stood out: access to real-time knowledge of the world through X, and a willingness to answer questions that most other systems decline.
Behind the chatbot was Grok-1, which xAI said it had developed over four months. An earlier prototype, Grok-0, had 33 billion parameters. Access at launch was a waitlist for a limited number of users in the United States, and the company called the release the first step of a longer roadmap.
The benchmark claims, and how xAI qualified them
xAI published a comparison table at launch. All figures were produced by xAI, and it noted that public benchmarks may have leaked into training data. To test for that, it hand-graded Grok-1, Claude 2, and GPT-4 on the May 2023 Hungarian national high school math exam, which was published after the training data were collected.
| Test | Grok-1 | GPT-3.5 | Claude 2 | GPT-4 |
|---|---|---|---|---|
| GSM8k (grade-school math) | 62.9% | 57.1% | 88.0% | 92.0% |
| MMLU (multiple choice, many subjects) | 73.0% | 70.0% | 75.0% | 86.4% |
| HumanEval (Python coding) | 63.2% | 48.1% | 70% | 67% |
| MATH (competition-style problems) | 23.9% | 23.5% | not reported | 42.5% |
| Hungarian exam, hand-graded | 59% | 41% | 55% | 68% |
The pattern is plain. Grok-1 beat GPT-3.5 on each test shown, if only narrowly on MATH, and trailed GPT-4 on every one. Claude 2 scored higher on all three public benchmarks where it was reported, but lower on the hand-graded exam. xAI stated that models trained with far more data and compute, GPT-4 among them, were ahead. The same page adds that, like other models trained to predict the next token, Grok can generate false or contradictory information even with search tools attached.
Open weights in March 2024
On March 17, 2024, xAI released the weights and architecture of Grok-1 under the Apache 2.0 license. It described the release as the raw base model checkpoint from the pre-training phase that ended in October 2023, with 25% of the weights active on a given token. Because it was not fine-tuned for dialogue, the release was not the chatbot people used on X. It covered Grok-1 only.
Timeline since the launch
| Date | Event |
|---|---|
| November 3, 2023 | Grok announced; limited early access in the United States |
| March 17, 2024 | Grok-1 base weights released under Apache 2.0 |
| July 12, 2025 | xAI apologizes for violent and antisemitic posts from the @grok account on X |
| January 14 and 15, 2026 | Privacy Commissioner of Canada widens its X Corp. investigation and opens one into xAI |
| June 11, 2026 | Commissioner reports that X Corp. and xAI violated PIPEDA |
| July 8, 2026 | Grok 4.5 released by SpaceXAI, as reported in news coverage |
| September 21, 2026 | Grok 4.7 released after several delays, as reported in news coverage |
The July 2025 apology
In its apology of July 12, 2025, xAI said the root cause was an update to a code path upstream of the @grok bot, separate from the underlying language model. The update, active for 16 hours, made the bot susceptible to existing X user posts, including posts with extremist views. xAI said it removed the deprecated code, rebuilt the system, and would publish the bot's new system prompt in a public repository.
What Canada's privacy regulator found
The Office of the Privacy Commissioner of Canada announced on January 15, 2026, that it was expanding an investigation into X Corp., first opened on February 27, 2025, over how Canadians' personal information was used to train AI models. The expansion followed reports that Grok was producing sexualized deepfakes. The Commissioner also opened a related investigation into xAI. Both asked whether the companies met their duties under the Personal Information Protection and Electronic Documents Act (PIPEDA), including valid consent.
The report of findings, issued June 11, 2026 (PIPEDA Findings 2026-004), concluded that Grok's image-generation tool was launched without proper safeguards or sufficient consideration of privacy harms. The Commissioner found breaches of the consent principle and of the appropriate-purposes provision in subsection 5(3), because depicted people had not given express consent and a reasonable person would consider sexualized deepfakes inappropriate. The tool, Grok Imagine, arrived in July 2025, and the privacy impact assessment that covered it was updated only in March 2026.
The companies added safeguards and proactive sweeps during the inquiry, which the Commissioner called encouraging while asking for more evidence that they work. They disagreed with the conclusions and with a recommendation to suspend Grok Imagine, but agreed to the other recommendations and to quarterly reports. The Commissioner also recommended an annual independent audit of safeguards, and the release urged lawmakers to add penalties and order-making powers to the federal law. The OPC will keep monitoring, and no later enforcement step was found as of October 2026.
Where the product stands in October 2026
News coverage reported that Grok 4.7 launched on September 21, 2026, after delays dating from late July. The article describes a model with 2.1 trillion parameters, priced at USD 2 per million input tokens and USD 6 per million output tokens, and quotes xAI saying it improves on Grok 4.6 at the same price and speed. Elon Musk described Grok 5 as a possible frontier leader without giving a date, and also mentioned Grok 4.8 and 4.9. These are statements of intent, not releases. The launch post came from an account named SpaceXAI, which matches July news coverage. Releases come quickly, so this section dates fast.
FAQ
When was Grok announced?
xAI's announcement is dated November 3, 2023. It was an early beta with a waitlist for a limited number of users in the United States.
Is Grok open source?
Only the Grok-1 base model is. xAI released its weights and architecture under Apache 2.0 on March 17, 2024. The release did not include the chatbot or later versions.
What did the Privacy Commissioner of Canada decide?
On June 11, 2026, the Commissioner found that X Corp. and xAI violated Canada's federal private-sector privacy law with the Grok image tool. The companies committed to quarterly and audited reporting.
Who makes Grok now?
xAI created it. News coverage on July 8, 2026 of Grok 4.5 calls the company SpaceXAI and describes X as its subsidiary. A September 2026 news report on Grok 4.7 still uses the name xAI for the developer.
What is the latest Grok model?
Grok 4.7, released September 21, 2026, per news coverage. Later versions may exist by the time of reading, and xAI's own news page could not be accessed for checking.
Related reading
- Inside the Grok conversational prototype
- Elon Musk's stated views on AI
- Cohere's enterprise funding round
- The ChatGPT assistant for comparison
More in AI news.
Last verified: October 2026
Sources
- xAI, "Announcing Grok," November 3, 2023
- xAI, "Open Release of Grok-1," March 17, 2024
- NBC News, "AI chatbot Grok issues apology for antisemitic posts," July 12, 2025
- Office of the Privacy Commissioner of Canada, news release on the expanded X investigation, January 15, 2026: source page
- Office of the Privacy Commissioner of Canada, news release on findings in the Grok investigation, June 11, 2026: source page
- "SpaceXAI releases Grok 4.5," July 8, 2026
- "xAI Launches Grok 4.7. It's Bigger, But Late to the AI Frontier Party," September 2026
- Office of the Privacy Commissioner of Canada, PIPEDA Findings 2026-004, June 11, 2026