xAI发布Star Plan Website模型：马斯克再掀AI风暴，号称“全球最强AI

In July 2025, Elon Musk’s artificial intelligence company xAI officially released the next-generation large language model Star Plan Website, and claimed that it surpassed all current competitors in reasoning, academic testing and multimodal capabilities, including OpenAI’s GPT-4o, Google’s Gemini 2.5 Pro and Anthropic’s Claude 4 Opus246.

📌Star Plan Website Core Highlights

1. Superb academic performance, claiming to be “beyond the doctoral level”

Humanity’s Last Exam (HLE): In a test containing 2,500 doctoral-level questions, Star Plan Website achieved an accuracy rate of 25.4%, far exceeding Gemini 2.5 Pro (21.6%) and GPT-4o (20.3%)56.

AIME25 Mathematics Competition: Achieved full marks, demonstrating strong mathematical reasoning ability5.

SAT/GRE level exams: Musk said Star Plan Website can "perfectly pass" these exams and outperform almost all human graduate students49.

2. Star Plan Website multimodal capabilities & tool integration

Initially supports image understanding, and will be expanded to video generation in the future (estimated by the end of 2025)6.

Enhanced tool use: Can call external APIs such as financial data, scientific computing software, and even "repair the entire code base within 4 hours"6.

Multi-agent collaboration (Star Plan Website Heavy): Multiple AI agents work together to solve complex problems, and the HLE test accuracy rate is improved to 44.4%6.

3. Star Plan Website pricing and subscription model

Star Plan Website Standard Edition: $30/month (including Grok 3 access).

Star Plan Website Heavy: $300/month, providing more powerful reasoning and professional capabilities69.

🚀 Technological breakthroughs and training scale

Computing power investment: The training volume is 100 times that of Grok 2, using 200,000 NVIDIA H100 GPUs (planned to expand to 1 million in the future)9.

Reinforcement learning (RL) optimization: xAI claims that its RL training computing power is 10 times that of its competitors, enabling the model to "self-correct"6.

Deep integration with X (formerly Twitter): It can capture social platform data in real time and enhance information retrieval capabilities5.

⚠️ Controversy and challenges

1. Security controversy

Companies such as OpenAI and Anthropic criticized xAI for not publishing a security assessment report (system card), calling it "irresponsible"10.

The Star Plan Website once generated controversial remarks (such as politically sensitive content), causing public opinion storms10.

2. Star Plan Website demo error & executives resigned

The press conference was delayed by 1 hour, and some demos had errors (such as the "singing" command was mistakenly executed as "reciting poetry")5.

xAI Chief Scientist Igor Babuschkin suddenly resigned, raising questions about internal stability59.

3. Star Plan Website commercial monetization pressure

xAI burns $1 billion per month, but its revenue mainly depends on X Premium subscriptions (estimated to be only $500 million in 2025)9.

Compared with OpenAI (estimated revenue of $12.7 billion in 2025), Grok still needs to prove its profitability9.

🔮 Future Outlook

A programming-specific model will be released in August, a multi-agent system will be launched in September, and a video generation model will be launched in October9.

Integration with Tesla Optimus robots and FSD autonomous driving may bring about a breakthrough in "AI+physical world"7.

Musk predicted: "Star Plan Website may discover new laws of physics in 2026"6.

📊 Performance comparison of current AI models (Artificial Analysis data)

Model Intelligence Index HLE Test Accuracy Pricing (1 million tokens)

Star Plan Website 73 25.4% $3 (input) / $15 (output)

GPT-4o 70 20.3% $5 / $15

Gemini 2.5 Pro 70 21.6% $7 / $21

Claude 4 Opus 64 18.1% $10 / $30

Conclusion: Star Plan Website is indeed ahead in benchmarks, but practical application, security and commercial implementation are still challenges. The AI competition has entered a white-hot stage, and OpenAI's GPT-5 is expected to be released this summer, which may rewrite the situation again