Qwen Glossary
Qwen (通义千问, Tōngyì Qiānwèn) is the open-weights large language model family released by Alibaba Cloud’s DAMO Academy, the research arm of Alibaba Group. The first Qwen generation was released in mid-2023; subsequent generations (Qwen 1.5, Qwen 2, Qwen 2.5, Qwen 3, Qwen 3.5, Qwen 3.6) have shipped at a faster cadence than most peer open-weights labs, with each generation expanding the parameter range from small (~0.5B) to large (~110B and beyond). The 2026-era flagship for local Apple Silicon use is Qwen 3.6 30B A3B — a Mixture-of-Experts model with ~30 billion total parameters and ~3 billion active per token, optimised for the kind of long-context tool-using agentic work the Dictionary’s audience runs.
For this Dictionary, Qwen matters at two levels. First, technically: the model is the current local-compute-tier leader on long-context-recall benchmarks (the kind of test where Gemma 4’s sliding-window attention pattern starts to hurt). The operator’s TOOLS.md routing rule pairs Qwen with large-context multi-file tasks, while Gemma 4 covers moderate-context contained tasks. Second, strategically: Qwen is the most-distributed example of a PRC-produced open-weights frontier-adjacent model. Its sustained competitiveness with Western open-weights families is one of the structural facts the zhengming (正名) project takes as load-bearing.
See also
- Open Weights
- Using the Empire’s Telescope
- Sovereign Compute
- Mixture of Experts
- DeepSeek — PRC peer
- Gemma, Hermes, Llama — local-compute peers