In the 2026 Spring Festival period, domestic AI large models delivered an impressive performance in the global developer ecosystem. According to AIbase's weekly data observations on OpenRouter (the world's largest large model API aggregation platform), the total token consumption of domestic models among the top ten on this platform has reached as high as 61%, demonstrating strong dominance by occupying the top three positions.
The top three models that "dominated the rankings" are MiniMax M2.5, Kimi K2.5, and Zhipu GLM-5. Among them, MiniMax M2.5 topped the list with an astonishing call volume of 24.5 trillion tokens. Developers generally believe that the breakthrough of domestic models in early 2026 mainly stems from technological advances in two key application scenarios: **codingandagents**.
As AI applications shift from simple "question-and-answer" interactions to complex automated workflows, the market demand for long text processing and low-latency responses has surged. Moonshot's Kimi K2.5 improved the efficiency of complex tasks by 3 to 10 times through the parallel collaboration of up to 100 "Agent avatars"; while Zhipu GLM-5 attracted a large number of developers handling ultra-long tasks, thanks to its 200K context window and sparse attention mechanism.
In addition to performance improvements, the extreme cost-effectiveness remains the "killer feature" that makes domestic models stand out. For example, the input cost of MiniMax M2.5 and GLM-5 is only 0.3 dollars per million tokens, while the price of the overseas flagship model Claude Opus 4.6 is as high as 5 dollars. AIbase believes that with a price advantage of about 16.7 times and comparable business understanding, domestic models are rapidly capturing market share in the global AI productivity market in 2026.
