Summary of the Core Content in One Sentence
This news article delves into the most crucial “internal rivalry” within the global large-model industry: OpenAI’s newly released GPT-6 Astra has regained the title of “world’s smartest model” in a standardized industry-wide performance test. However, in the past year, Anthropic, founded by former key employees of OpenAI, has surpassed its former employer in terms of market share, annual revenue, and valuation by focusing on the code development scenarios that businesses are most willing to pay for. The competition between the two companies has moved beyond the initial stage of simply comparing model performance scores and has shifted to a more substantial focus on integrating AI into daily business operations to generate long-term, stable revenue. Leading in model performance alone is no longer enough to win the commercial battle.
---
Breakdown in 5 Easy-to-Understand Points
1. OpenAI’s “First Place” Is Only About Test Scores
Many people think that OpenAI has regained the market dominance when it claims to have the world’s most intelligent model, but the significance of this title should be considered more cautiously:
The so-called scores of 97.6% in mathematics, 99.9% in common sense reasoning, and 100% in hacking defense tests are based on a standardized set of questions used across the industry. Astra scored nearly full marks on these tests, reversing the situation where Anthropic’s Claude series had dominated for over half a year. This means Astra has regained its technical reputation. However, this does not equate to market dominance: In independent third-party rankings, Astra is now on par with Anthropic’s latest model. High test scores do not necessarily mean that users are willing to pay for the service; it’s like a new phone having 20% better performance than competitors, but whether it can sell better is another matter entirely.
2. How Did Anthropic Overcome OpenAI?
The founder of Anthropic was once OpenAI’s second-in-command in research. In 2020, he led a group of core employees to start their own company and has been catching up with OpenAI for several years until they finally overtook them last year. Instead of competing for the general consumer market, Anthropic targeted the most mature commercial application: coding. Silicon Valley companies spend $37 billion on AI, of which $4 billion is specifically for coding tasks. Anthropic has dominated this segment, holding 40% of the total AI API market share, while OpenAI only has 27%. This revenue gap is significant—Anthropic could earn $65 billion annually compared to OpenAI’s $40 billion, and its valuation has reached $965 billion, surpassing OpenAI’s $852 billion. Even in China’s venture capital community, there is a search for a “Chinese version of Anthropic” as a benchmark for large-model commercialization.
3. OpenAI’s Competition Is About Access to Business Operations
OpenAI’s release of Astra is not aimed at ordinary ChatGPT users but at the areas where Anthropic has a strong presence. OpenAI emphasizes its ability to help with coding, operate software on users’ computers, and complete tasks from start to finish without human intervention. It is following the path Anthropic has already established. This year, OpenAI launched an Agent platform that can access internal company data and systems and partnered with top consulting firms like McKinsey, PwC, and Accenture to expand its customer base. Its Codex tool has 15 million active users per week, including many non-programmers such as operations staff and analysts. The goal for both companies is to integrate AI directly into business processes, as the money in the large-model industry comes from providing practical solutions, not just from being chat assistants.
4. New Rules in the Large-Model Industry: High Scores Don’t Equal Real Revenue
Many wonder why companies still use Anthropic’s model despite its lower scores. The logic behind companies’ AI purchases is different from that of individuals buying phones. While individuals might spend more for a new phone with better performance, companies don’t necessarily pursue the latest models. Data shows that only 6% of companies use Anthropic’s latest, most powerful model; 94% prefer older models with better cost-effectiveness. Once the top models can perform 80% of the required tasks, companies are less willing to pay significantly more for the remaining 20%. They care more about security (storing data on their servers) and seamless integration with existing systems. These factors are far more important than model performance. Even if OpenAI’s model scores are higher, it won’t easily attract Anthropic’s existing customers.
5. The Next Round of Competition: Integrating AI into Business Processes
The next stage of competition is not about which model is smarter but about who can effectively integrate AI into various business processes. In the future, companies will use multiple AI systems for different tasks (coding, reporting, customer management, design, etc.). The key will be to enable these systems to work together seamlessly without human intervention. OpenAI is developing a platform to manage all internal AI systems, while Anthropic is working on a universal protocol for connecting different AI systems. The company that can successfully integrate these technologies will be the true leader. By then, the current focus on model scores will be of little significance.