Exclusive | Zhipu's ARR Hits $1 Billion, Surges 15x in Six Months

Article | Xinyu Zhou

Editor | Yuxin Zhang

Intelligent Emergence has learned from multiple independent sources that as of July 2026, Zhipu's ARR (Annual Recurring Revenue) has reached 1 billion USD.

As of press time, Zhipu has not responded to the aforementioned information.

Over the past year, AI coding and video generation models have become the AI tracks with the strongest self-sustainability capability globally.

Overseas, Anthropic's Claude Code saw its ARR surge to 1 billion USD just half a year after launch. This cash cow product also lifted Anthropic's overall valuation to 1 trillion USD (in May 2026), surpassing OpenAI.

However, the market has long held doubts about whether Chinese large model companies can replicate Anthropic's wealth myth. On one hand, backed by the North American market with high enterprise payment willingness and a mature subscription ecosystem, Anthropic enjoys a naturally favorable commercialization environment. On the other hand, there is still an order-of-magnitude gap between Anthropic's valuation and Zhipu's latest market capitalization — meaning Anthropic has far more funds to access training resources, a advantage that domestic model manufacturers cannot match yet.

But the speed at which Zhipu reached 1 billion USD in ARR has far exceeded expectations. An investor once told us that Zhipu's ARR was projected to reach 1 billion to 1.5 billion USD only by the end of 2026.

A more intuitive comparison: According to official Anthropic data and Zhipu's financial reports, it took Anthropic 15 months to grow its ARR from 100 million USD to 1 billion USD, while Zhipu only spent 5 months. According to insiders, between January and July 2026, Zhipu's ARR increased 15-fold.

Zhipu is one of the earliest domestic model manufacturers to bet on the coding path. Back in early 2025, Zhipu began to concentrate its efforts on enhancing the model's coding capabilities. Tang Jie, the founder of Zhipu, once described that Zhipu has bet on coding and reasoning — a model capability that can co-evolve and thrive alongside AI agents.

Subsequently, Zhipu launched a series of flagship models with coding capabilities ranking in the global top tier at an average pace of one new version every two months. The latest GLM-5.2 launched in June 2026, even as an open-source model, still matches or even outperforms Claude Opus 4.8 and GPT-5.5 in multiple core metrics.

Zhipu's official team once mentioned that the commercial value of large models can be defined as "Intelligence ceiling × Total token consumption volume": as long as the model's capabilities maintain a leading edge, the usage volume will keep growing. Financial reports show that in Q1 this year, the cumulative price of GLM's APIs was raised by approximately 83%, and the overseas subscription price is nearly on par with Claude Code, yet the usage volume still increased by around 400%.

At a time when token consumption is used to measure the commercialization level of models, coding and video generation models represent two distinct paradigms: the former directly delivers productivity to enterprises and developers, while the latter penetrates into the core links of industrialized content production.

In terms of revenue data, the market demand for coding has become clearly defined, and so has that for video generation. According to Intelligent Emergence, Seedance 2.0's monthly revenue is nearly 1 billion RMB; Goldman Sachs also predicts that Kling AI's ARR will reach 1 billion USD by the end of the year.

However, a clearly defined business also means more intense competition. In June, MiniMax released M3, with a focus on strengthening coding and agent capabilities. On July 16th, Moonshot AI also launched K3, an open-source model with 2.8 trillion parameters, whose overall intelligence level is second only to Claude Fable 5 and GPT-5.6 Sol.

Even overseas, with the release of GPT-5.6 and the integration of ChatGPT and CodeX, OpenAI has shown a momentum of overtaking Anthropic in the coding domain.

At this current point in time, the battle for coding dominance is far from over.