Bonsai 27B Slammed as Nearly Unusable
AdamLangePL · reddit · 2026-07-15
A post discusses the poor performance of Bonsai 27B on simple tasks. The author claims it is "almost useless" for coding, often truncating its thought process and failing to provide a final result.
The author downloaded the quantized model from the PrismML HF page, ran it using the llama.cpp llama fork, and provided two examples:
- Asking the model to "tell a story"
- Writing a simple Cfunction
The author also adjusted the context length to the model's maximum limit, but the results did not improve.
More from Models
- Local benchmark: Flint matches Qwen3.6 quality at 50/52 while running 3.6x faster on M4 Pro — RichardSocher · 2026-10-27
- ICML 2026 position paper: LLM 'thinking traces' are a harmful anthropomorphism — rao2z · 2026-10-27
- One stray 'u' and Claude starts spelling like a Brit — cantrell · 2026-10-27
- Cohere's Invent-A-Dataset paper: zero-seed synthetic data beats frontier APIs on quality and diversity — idanbeck · 2026-10-27
- ARC Prize redesigns leaderboard: highest reasoning effort by default, CSV export — mhmazur · 2026-10-27
- Cohere: open-weight Qwen3.8-Flash-Next tried uploading probe file to Beijing cloud bucket in sandbox — wstone_bd · 2026-10-27