Musk announces rollout of Grok 4.5, says it’s as good if not better than Anthropic’s Claude Opus

Grok 4.5 has been released for private beta testing, and Elon Musk is confident that this model’s performance is on par with, or potentially exceeds, Anthropic’s flagship model.

Musk has revealed new details surrounding xAI’s most powerful AI model to date.

Posting on his social media platform X, Musk said that Grok 4.5, which is trained on xAI’s 1.5T V9 foundation model, is being tested internally at his companies, SpaceX and Tesla.

Musk has released Grok 4.5 in a private beta, meaning it will be tested within his companies to fine-tune bugs and potential issues with the model until it’s ready for the public.

xAI uses a machine learning technique called reinforcement learning, which mimics the human trial-and-error method, aiming to develop AI models that make the best decision possible based on the prompt.

This method, coupled with xAI’s coding agent Grok Build, should “significantly improve the model,” as both Grok 4.5 and Grok Build get “better every day,” according to Musk’s post.

Musk also mentioned that Grok 4.5 used Cursor data, which was added in supplemental training.

Cursor, the AI coding platform acquired by Musk for $60 billion, was used to help train Grok 4.5 later on in the model’s development.

Early evaluations of Grok 4.5 show that the model shows that its performance is “close to (or) perhaps exceeding Opus,” Musk claims.

Opus, which is used by major companies like Shopify and Cursor, is typically used for advanced coding tasks and knowledge work, according to Anthropic.

The flagship model is regarded as Anthropic’s “most capable Opus-tier model for complex reasoning and agentic coding,” and is considered the company’s most powerful consumer model.

Anthropic introduced Opus 4.8 a month ago, and one notable revision to the model involved its capacity to be honest.

“One of the most prominent improvements in Opus 4.8 is its honesty. We train all our models to be honest, for instance, to avoid making claims that they can’t support,” said Anthropic.

As AI models are pattern recognition machines and don’t have context of the real world, they tend to “jump to conclusions” and make convincing claims, while there’s limited evidence to support their findings.

Anthropic’s internal evaluations measured for factual hallucinations and found that Opus 4.8 had “the lowest incorrect-rate” of all 6 models.

Opus 4.8 achieved this by refusing to answer questions it was uncertain about rather than answering more questions correctly,” Anthropic found.

Grok was developed to rival OpenAI’s ChatGPT and was designed to be a rebellious yet honest truth-teller.

While various scandals, including Grok’s praising of Adolf Hitler, “white genocide in South Africa, Grok creating deepfake nudes, government bans, and casual use of Hindi slurs, tainted the model’s development, xAI continued to assert that Grok is honest and is less likely to hallucinate than its previous models.

Musk’s comparison of Grok 4.5 to Opus suggests that the private beta model will follow similar principles to Anthropic.

Musk’s perpetual pursuit of honesty

While Musk’s products, namely Grok, have been scrutinized by lawmakers and the public for many of the reasons featured above, honesty is seemingly a core principle of Musk’s personal and professional brand.

An independent developer, Musk superfan, and creator of the account Muskosophy has positioned the tech billionaire as a painfully honest person.

In an X post, Muskosophy shared Musk’s philosophy on honesty and truth, saying that he believes that objective truth and honesty are required to unlock the mysteries of the universe.

“If you're not rigorous about truth and honesty, you're going to live in a deluded world, and you won't understand the nature of reality,” Musk said.

The SpaceX CEO responded to this post, saying that his attitude towards life “is the only way.”

This attitude is mapped throughout most of Musk’s products, with X positioned as a “digital town square” where users can fact-check posts using Community Notes.

However, X has long been criticized as a cozy bubble for right-wing conservatives, with studies reinforcing the argument that X algorithms prioritize conservative content and sway opinions towards the political right.

Musk also created Grokipedia, an alternative to Wikipedia, which also claimed to promote honesty and transparency when looking for information online.

Grokipedia is positioned as “the world’s largest and most accurate knowledge source without centralized control” and was created to solve issues within Wikipedia’s editorial biases.

DogeDesigner, which almost functions as an Elon Musk fan account, claims that Grokipedia is an antidote to Wikipedia’s left-wing bias and claims the platform is “often used as a propaganda tool, not an unbiased encyclopedia.”

However, Grokipedia uses Grok to decide what information to include and what data is correct or incorrect.

Because many of Musk’s products are based on Grok, which is trained on a wide variety of data, including X posts, the former trillionaire’s perilous pursuit of truth may be stifled by misinformation and far-right ideologies proliferating across his platform.

Most recently, former xAI engineer Devin Kim was fired from the company after pushing for Grok to be developed with stronger safeguards.

xAI failed to prioritize AI safety when developing Grok, which “virtually guaranteed that the company would commit unlawful acts.”

Without these safeguards, Grok would likely encourage discrimination and teach users how to develop weapons of mass destruction.

---

Unlock more exclusive Cybernews content on YouTube.