Elon Musk, founder of SpaceXAI, made an unexpected admission: his flagship neural network Grok 4.5 is inferior to Anthropic's Claude Fable 5 model. In a post on platform X, he directly stated that Fable performs better, although he emphasized that Grok 4.5 remains extremely fast, cost-effective, and handles most everyday tasks excellently.

This statement was a response to a user who claimed that Grok 4.5 performs at the level of Fable but costs significantly less. Musk decided to clarify: yes, his model is cheaper, but in pure quality, it falls short of Anthropic's flagship.

What exactly Musk said and why it matters

Admitting to lagging behind a competitor is an extremely rare move for the head of a major technology company. Musk added that for most everyday tasks, such a high level as Fable's is not required, and Grok 4.5 remains the optimal choice in terms of price/quality ratio. He had previously noted that on some software tests, Grok 4.5 still outperforms strong competitor models.

The market reacted cautiously to these statements, indicating that investors understand SpaceXAI's long-term strategy: a focus on accessibility and speed, rather than absolute leadership in every benchmark.

Comparing Grok 4.5 with competitors

SpaceXAI developers have bet on cost and data processing speed. The Grok 4.5 model costs only $2 per million input tokens and $6 per million output tokens. For comparison, competitor rates are significantly higher:

AI ModelInput per 1M tokensOutput per 1M tokens
Grok 4.5$2$6
Claude Opus 4.8$5$25
GPT-5.6 Sol$5$30

Independent tests confirm the price advantage with some quality lag. In the Artificial Analysis intelligence index, the Grok 4.5 model ranked fourth with 54 points. The list leaders remain Fable 5 with 60 points, Opus 4.8 with 56, and GPT-5.5 with 55 points. In applied tests, results were also mixed: on SWE Marathon, Grok 4.5 outperformed all with a score of 29%, but on SWE Bench Pro, it scored 64.7%, falling behind Claude Opus 4.8 (69.2%) and Claude Fable 5 (80.4%).

  • On the agent test AutomationBench-AA, the Grok 4.5 model scored 51.4%, surpassing Claude Fable 5 (48.6%) and Claude Opus 4.8 (48.5%).
  • The cost of solving a task with Grok 4.5 was only $0.34 compared to $1.35 for Fable 5 and $1.46 for Opus 4.8.
  • It is this productivity lead that Musk previously used as the main proof of his network's strength.

However, such efficiency also has a downside. Experts point to a higher frequency of rule violations as the model's weak point. This creates additional risks in corporate deployment of AI agents. The main point of this release boils down to simple economics: developers show results where a strong price difference outweighs a slight lag in tests.

Expert opinion: Musk's admission is not a weakness but a calculated marketing move. He clearly positions Grok 4.5 as a "people's" model for mass tasks, rather than a premium tool for niche specialists. In the long term, this strategy could bring SpaceXAI a larger market share than chasing abstract records in benchmarks.