Zhipu AI Announces Next-Gen LLM Strategy and H1 2026 Financial Results
Zhipu AI announced that its next-generation large language model will adopt a "large base" route, focusing on mid-training and post-training to support full-scale business deployment on day one. Additionally, the company reported H1 2026 revenue of 954 million RMB, representing a 399.7% year-over-year increase, while narrowing its net loss to 2.071 billion RMB. This strategy addresses a major industry bottleneck where massive models are released but remain difficult to deploy practically due to inference constraints. Zhipu's explosive revenue growth highlights the rapid commercialization of generative AI in China, even as high development costs keep the company unprofitable. While specific parameter sizes for the next-generation model remain undisclosed, Zhipu AI is already conducting inference-side pre-research to ensure immediate scalability. The company's gross profit for the first half of 2026 reached 252 million RMB with a gross margin of 26.4%.
## BACKGROUND
Zhipu AI is a prominent Chinese AI startup known for its GLM (General Language Model) family, which includes models like GLM-4 and GLM-5. In LLM development, "mid-training" is an intermediate phase between pre-training and fine-tuning that refines model capabilities (such as reasoning or coding) using high-quality curated data.