SpaceXAI’s “most capable model” is really average and has more guardrails
Independent authority shows Grok 4.7 is bang average.
Grok logo with flares. By Nur Photo/Getty Images.
- SpaceXAI announced its latest version of Grok claiming its much better than the 4.6 model.
- Independent analysis places Grok 4.7 below leading models from Anthropic and OpenAI while securing only 3 more points than Grok 4.6.
- SpaceXAI says Grok 4.7 adds stronger safeguards against prompt refusals and jailbreak attempts.
- Some users say Grok has lost its main appeal after adding stronger content restrictions.
Key Takeaways by nexos.ai, reviewed by Cybernews staff.
SpaceXAI has released its latest version of Grok, boasting enhanced coding and knowledge capabilities at the same price. But, is it as impressive as they claim?
Grok 4.7 is SpaceXAI’s “most capable model” to date, with better performance in coding and knowledge work than its predecessors, at the same price and speed as Grok 4.6.
Elon Musk has said that Grok 4.7 is “significantly faster and lower cost,” while saying it’ll compete with Anthropic’s Opus 5.
While Grok 4.7 has helped SpaceXAI achieve 3rd place in the global AI lab rankings, the latest model only scored a few points higher than 4.6 and didn’t come close to matching Claude Opus 5.
Despite Musk’s attempts to hype Grok 4.7 ahead of its release, the model is mid-tier and compares more closely to Meta’s Muse Spark and Qwen 3.8, according to independent analysis.
Is there anything special about Grok 4.7?
Grok 4.7 isn’t much better than the previous model, gaining just 3 points (from 43 to 46) on the Artificial Analysis Index, which is an independent evaluation authority for AI models.
In comparison to other models, Grok 4.7 lags behind top players like Anthropic’s Fable 5.1 (max with fallback) and OpenAI’s GPT-6 Astra (xhigh). Both of these models lead in intelligence, with 53 points each.
The capabilities of this latest version do compare to GPT-6 Astra and Claude Fable 5.1, but only when both models are performing at their lowest possible level, while Grok 4.7 performs at extra high and high.
However, this version is arguably more efficient, as it operates more quickly and at a lower cost per task.
When SpaceXAI announced Grok 4.7, it said it was “built with an entirely new safeguard stack” and is the most competent model in terms of prompt refusals and jailbreak resistance.
Despite Musk’s initial promise that SpaceXAI’s model would be unfiltered and raw, the tech billionaire seems to be prioritizing user safety more than we might have thought.
Regardless, Grok 4.7 isn’t living up to Musk’s optimistic expectations, and users know it.
Stay updated with our latest stories and follow us on social media
Be the first to discover new stories, ideas, and updates from our team.
Grok is useless, say users
Given that Grok was designed to be a free-speech bot capable of generating outputs other AI companies wouldn’t dare imagine, the fact that Musk is heavily fortifying its AI model leaves users asking, “What is Grok even used for?”
On Reddit, users claim that Grok is actually worse than Meta’s AI model, Gemini, a divisive topic among users.
In response, users seem to suggest they used Grok only for “spicy images” and “spicy LLM prompts,” not because it is a better model than those produced by Anthropic or OpenAI.
“Grok sucks across the board. It doesn’t do NSFW images anymore, and the guardrails are way worse. It’s a pointless model,” declared one Reddit user.
Users echoed the sentiment, saying that Grok was superior when it offered something that other AI labs didn’t.
“Grok was useful when it prided itself on being the no-censorship model. That, of course, had terrible consequences, morally, socially, and, which is likely why they changed strategies, legally. Nowadays, it simply serves no purpose.”
These “terrible consequences” are likely attributed to the Grok pornography scandal, where Grok was capable of creating non-consensual deepfake pornography of women and children.
Since then, Grok has been sued by teenagers who allege the AI model made child porn of them, has sued users for creating child pornography, and has been accused of training Grok on child pornography.
