A quantized version of the Qwen3.8-27B model has been released

Developers introduced Qwen3.8-27B-GGUF — an optimized version of the language model, adapted for local execution on consumer equipment. The release includes quantized weights, allowing the model to be used with limited hardware resources without significant loss of accuracy.

The model demonstrates a result of 85 points in internal tests, achieving this indicator in 1 hour of data processing. The GGUF format ensures compatibility with popular runtime environments, including llama.cpp, vLLM and Ollama. The use of quantization allows running the 27-billion model on systems with less video memory than required to run the full-size version in FP16 format.

The toolkit is aimed at developers implementing local AI solutions in applications like Pi or Hermes. Support for the GGUF format simplifies the integration of the model into existing local inference pipelines, providing high token generation speed on standard GPUs and CPUs.

This release expands the availability of the Qwen architecture for enthusiasts and companies working with private data. The possibility of local deployment eliminates the need to transmit requests to external servers, which is critical for tasks with increased information security requirements.