Term
Qwen 2.5
Open model family from Alibaba, released in September 2024. It spans base sizes from 0.5B to 72B parameters plus Coder and Math variants; 128K context, multilingual. Smaller sizes under Apache 2.0, the 72B model under a dedicated Qwen license.
Qwen 2.5 — explained in more detail
Qwen 2.5 is an open model family from the Chinese company Alibaba, released in September 2024. It is regarded as an important step for open-weight language models because it shipped in a broad range of sizes with permissive licensing. The weights can be downloaded, run locally and fine-tuned.
The family comprises seven base sizes: 0.5B, 1.5B, 3B, 7B, 14B, 32B and 72B parameters. In addition, there are specialised Qwen2.5-Coder (trained on program code) and Qwen2.5-Math variants, plus quantised editions (GGUF, AWQ, GPTQ). All models offer a 128,000-token context window with up to 8,000 tokens of output and were pretrained on roughly 18 trillion tokens. Qwen is also strongly multilingual.
On licensing, the sub-72B sizes are released under Apache 2.0 with no usage cap, while the 72B model is under the dedicated Qwen license (with conditions above very high user counts).
Example / practical relevance
The range of sizes makes Qwen 2.5 flexible: small models (0.5B–3B) run on weak hardware or on-device, while large ones (32B, 72B) deliver more quality for demanding tasks. The Coder variant suits code generation and analysis, the Math variant compute-heavy tasks.
For teams with data-protection or cost requirements, the Apache 2.0 license of the smaller models is attractive: self-host, use freely, with no per-request API cost.
Distinction from similar terms
Qwen is Alibaba’s model family and stands alongside other open families such as Meta’s Llama or Mistral. Qwen 2.5 is an older generation; later versions (Qwen 3 and beyond) followed with mixture-of-experts architectures. The “Coder” and “Math” suffixes denote task-specialised offshoots of the same generation, not separate model families.
Discover more
GPT-6 Astra Found Questions My First Security Review Missed
I used GPT-6 Astra and Fable 5.1 as independent reviewers for RLS, API and tenant-isolation checks. The useful part was the structured cross-review.
GlossaryQwen 3.8-Max
Qwen 3.8-Max (August 2026) is Alibabas largest flagship — an MoE model with 2.4T parameters, a 1M-token context and multimodal input, later released with open weights.
EncyclopediaComparing AI models — who builds what and how to choose
The major model families in 2026 at a glance. Who builds Claude, GPT, Gemini, Llama, Mistral, DeepSeek, Qwen — and which model to pick when.