Qwen2.5: A Party of Foundation Models!

Qwen's Qwen2.5 launch offered a family of models tailored to different workloads. Alongside general language models, it released coding and mathematics specialists. The September 2024 announcement describes code-focused training for Coder and mathematical training that supports written reasoning and the use of programs or tools to solve problems. Developers could therefore evaluate a specialist for a defined task rather than choose solely by model size.

The general family ranged from 0.5B to 72B parameters, but release conditions varied. The announcement identified 3B and 72B licensing exceptions to Apache 2.0, and the 32B Coder was still forthcoming. The family name alone did not specify which capabilities or terms applied.

“We are announcing what might be the largest opensource release in history!”