27B ternary model runs locally with 262K context on regular PCs

27B ternary model runs locally with 262K context on regular PCs

Bonsai-27b is a ternary-quantized version of the Qwen3.6-27B model, compressing its weights down to just 1.58 bits. This extreme compression lets a 27-billion-parameter model run locally via Ollama on ordinary consumer hardware instead of requiring server-grade GPUs.

The model supports multi-step reasoning and , and offers a very large 262,000-token . Model files and are available for direct download from a repository.

Key points

  • Ternary (1.58-bit) quantized version of Qwen3.6-27B
  • via Ollama on consumer-grade hardware
  • Supports multi-step reasoning and
  • Offers up to a 262K-token
  • Model and downloadable directly from
Read original