Musk Says Anthropic Leads AI With No Rivals, Then Predicts xAI Will Overtake Everyone in Six Months

The admission

Elon Musk told X users that Anthropic's Claude models are the current leader in AI, calling out the company by name and saying no rival matches its top models, according to Crypto Briefing. He didn't hedge. He said he was wrong about where Anthropic stood in the field.

The numbers back him up. Grok 4.5, which xAI launched July 8, 2026, at $2 per million input tokens and $6 per million output tokens, scored just 53% on the DeepSWE 1.1 coding benchmark, according to Crypto Briefing, trailing Anthropic's Claude Fable and Opus line across multiple categories. Musk's own explanation: xAI has been building models for roughly three years. Anthropic has had six.

The delayed model

That gap is the backdrop for Grok 4.7, which missed its original September 12 target. Musk explained on X that the reinforcement learning process penalized response length too aggressively, causing the model to "give up on hard tasks (that it can do!) too early," according to posts reported by both Cryptopolitan and Digital Today. He said the model "should be roughly on par with Opus 5.0, not 5.1" and that multimodal performance still needs work.

Grok 4.7 was reported at roughly 2.1 trillion parameters, built with SpaceX engineering data folded in, about 40% larger than Grok 4.6's reported 1.5 trillion, per Cryptopolitan. By contrast, Anthropic's Claude Fable 5.1 was already public with a confirmed score of 52.6% on Terminal-Bench-Science and pricing of $10 per million input tokens and $50 per million output tokens.

Grok 4.7 did launch this week, and Musk ranked it third in agentic coding, behind Anthropic and OpenAI, according to StockTwits. Anthropic released Claude Opus 5.5 in the same stretch, and OpenAI added GPT-6 Sol and Luna. On Artificial Analysis's intelligence benchmark, cited by AI Magazine, Grok 4.7 scored 46 against Opus 5.5's 58 and OpenAI's GPT-6 Astra (max) at 53. Grok 4.7 ranked 16th overall on a broader benchmark, though it beat Alibaba's Qwen3.8 Max and Kimi K3.

The six-month prediction

On Thursday, September 24, Musk pivoted to confidence. He posted that he is "cautiously optimistic" his company will have a model comparable to Claude Fable or GPT-6 within two to three months, and could reach "pole position" in AI within about six months if its pace of improvement holds, according to Tekedia and AI Magazine.

His reasoning rested on three points. First, xAI's effort is only three years old against Anthropic's six and OpenAI's ten, so its improvement curve is steeper. Second, he argued raw intelligence has diminishing returns once a model clears the bar a task actually requires, illustrated with his line that it would be "cruel" to put "Newton-level intelligence" in a toaster. Third, he pointed to hardware: SpaceX's ability to bring massive compute online, including its Colossus data centers and a planned chip manufacturing project called Terafab in Texas.

SpaceX has a real track record of scaling industrial infrastructure faster than competitors expected, from Starlink satellite production to Starship's launch cadence. Musk's bet is that the same execution edge translates to compute. The claim itself comes with no benchmark data and no release date, a gap AI Magazine and BigGo Finance both flagged directly. The measurable baseline right now is a model, Grok 4.7, that trails Opus 5.5 by 12 points on Artificial Analysis's own scale.

Markets, and a side fight over safety

SpaceX, which acquired xAI in February 2026 and began trading on Nasdaq as SPCX in June, is the closest public proxy investors have for this bet. The stock fell 0.2% Thursday, extending a weekly decline to 3% and snapping a four-week winning streak, then rose 0.2% overnight heading into Friday after Musk's six-month post, according to StockTwits. SpaceX's AI segment generated about $2.56 billion in second-quarter revenue.

A separate Musk post on September 14, endorsing AI slowdown warnings from Anthropic CEO Dario Amodei and OpenAI CEO Sam Altman, coincided with declines in Nvidia and AMD shares as investors weighed a potential pullback in AI infrastructure spending, according to BigGo Finance. President Trump called the safety push a "sick conspiracy," per the same report. No further detail on Trump's remark or an administration response was available in that reporting.

What comes next

Musk has also laid out a roadmap past 4.7: Grok 4.8, a 2.5-trillion-parameter model on a new C++ software stack, was finishing training and moving into reinforcement learning as of mid-September, which he called a "noticeable improvement." Grok 4.9 is projected to reach Astra or Fable-class performance. Grok 5, still without a benchmark, timeline, or spec sheet, is the model Musk has flagged as xAI's first attempt at artificial general intelligence, telling one user simply, "that will be Grok 5."

Notably, xAI's SpaceX-Anthropic relationship isn't purely adversarial. SpaceX has a compute agreement supplying Anthropic hundreds of megawatts of power from the Colossus facilities, and Musk has pledged to keep infrastructure access level rather than use his hardware position against a rival he just called the industry leader.

The question is whether Artificial Analysis or any other independent benchmark tracker will show Grok closing that 12-point gap on Opus 5.5 before Musk's six-month clock runs out in March 2027.