Recently, a domestic AI company listed on the Hong Kong stock market, claiming to have secured the title of "Global No.1 Large Model." At first glance, it's quite surprising, but after understanding their technological progress, it makes perfect sense.

This company has always been relatively low-profile. The real splash in the international industry came last September with a product update—the release of the GLM-4.6 version. Its coding ability increased by 27%, tying for first place in the authoritative Code Arena evaluation worldwide, becoming the first domestically developed model comparable to Claude Sonnet 4. Additionally, it optimized token consumption, reducing it by 30%. Industry insiders recognized this at the time, but the general public paid little attention, as programming tools are still somewhat niche.

However, the situation has changed over the past six months. Starting in the second half of 2025, the global development trend of large models shifted. No one is anymore touting universal AGI; everyone understands it won't be achievable in the short term. Moving into 2026, it’s clear that practical applications will be far more popular than theoretical breakthroughs. And the most promising application field? AI programming without a doubt.

Why? Because this can directly generate revenue. Claude focuses solely on programming. Although it invests far less resources than Google and OpenAI, it has become the most popular in AI programming. Their parent company, Anthropic, made an interesting estimate last October—that by the end of the year, their ARR (Annual Recurring Revenue) could reach $9 billion, with a baseline forecast of $20 billion in 2026. The Hong Kong-listed company managed to get ahead of the market with its GLM programming solution. Their developer-oriented programming plan has already surpassed 100 million RMB in ARR. Even more impressive—over 2.7 million paying customers, establishing a solid user base.

The industry now agrees: AI programming cannot rely solely on dataset scores; it must outperform in real-world applications by error rate and continuous execution stability. Interestingly, AI programming is like a "self-reinforcing" process for programs, as well as a "self-revolution" for programmers, carrying a sense of sci-fi and desolation, quite different from other AI applications. Currently, in the large programming model track, domestically developed GLM-4.7 is close to or even surpassing mainstream US models, which is no small feat. Coupled with domestic price and efficiency advantages, the competitiveness is evident.

Looking ahead, the promotion of AI programming applications in 2026 will surely make significant progress. More people will truly experience the power of these tools. The company that launched early has already gained a market share and application ecosystem advantage, laying a solid foundation in the domestic large model programming field.