Ternary Bonsai 2 27B

A 27B-parameter model from PrismML based on Qwen, compressed with ternary quantization to reduce size while maintaining reasoning capability across coding, mathematics, and tool use.

Strengths

Delivers reasoning performance for code and math tasks in a compact 27B footprint, supported by a 262K-token context window for handling longer problems and multi-step reasoning.

Best for

Projects where a mid-size model with strong reasoning, coding, and math abilities fits your inference and memory constraints, particularly those requiring extended context.

Limitations

Ternary compression introduces quantization trade-offs that may reduce fidelity on complex tasks compared to full-precision alternatives, and at 27B it sits between smaller and larger reasoning models.