Open-Source Model Wave Continues: DeepSeek, Qwen, and GLM Releases
View original source →This week continued the sustained wave of capable, freely available AI models primarily from Chinese research labs. These models are not niche experiments — they are being used in enterprise pipelines, security research, and scientific workflows.
Key releases:
• DeepSeek released DeepSeek-V4-Pro-0813 and DeepSeek-V4-Flash-Vision-Exp — advancing both reasoning and multimodal visual processing capabilities in open repositories • Alibaba's Qwen team released Qwen3.8-27B — a 27-billion parameter model targeting localized edge deployment and mathematical reasoning, practical for organizations wanting capable local AI without massive hardware • Zhipu AI released GLM-5.3 — focused on agentic tool invocation and multilingual task execution across enterprise use cases • Thinking Machines Lab published the open-weight Inkling foundation model for research experimentation and custom fine-tuning • xAI released Grok 4.6 continuing rapid iteration on its developer-focused model integrated into the Cursor IDE ecosystem
Combined, open-weight models from Chinese labs now account for approximately 30% of global open AI inference volume, actively depressing pricing power of closed proprietary APIs. The open-source AI ecosystem is advancing fast enough to narrow the capability gap with proprietary models quarter over quarter.
Why It Matters: Open-weight models are a practical procurement option now, not a future promise. The capability gap with proprietary models is narrowing quarter over quarter, and they offer options where closed US models are too expensive or too restricted.