Zhongjin: In 2026, significant breakthroughs will be made in reinforcement learning, model memory, contextual engineering, and other areas in the field of large models.

Zhongjin pointed out that looking back to 2025, global large model technology capabilities have been advancing, gradually overcoming productivity scenarios in reasoning, programming, Agentic, and multimodal abilities. However, there are still shortcomings in the stability and hallucination rate of model general capabilities. Looking ahead to 2026, Zhongjin believes that large models will make more breakthroughs in reinforcement learning, model memory, context construction, and other aspects, progressing from short-context generation to long-chain thinking tasks, from text interaction to native multimodal, and further towards achieving the long-term goal of AGI.