Ternary-quantized Bonsai-2 outshines Qwen3-VL as a tiny prompt-writing model
MehZhure · reddit · 2026-11-13
A Redditor shares a counterintuitive use case: the Bonsai models, often criticized for poor tool calling, turn out to excel as tiny prompt writers. Bonsai-2 is a distillation of Qwen3.8-27B available as a ternary quant — nearly matching the full 27B's writing quality at 9B-or-less size.
While testing mondo-music-video (an open-source tool that turns songs into beat-aware music videos with LLM-written prompts and local or hosted rendering), the author found the default qwen3-vl writer kept leaking thinking tags into output. Swapping in Bonsai-2 with thinking disabled worked great. Weights: prism-ml/Ternary-Bonsai-2-27B-gguf; project: hodgemann/mondo-music-video.
More from coding & agent
- Zhejiang U & Baidu's Agent-R²: 4B agent beats larger open-source baselines on GDPval — ZhejiangUniversity · 2026-11-13
- Claude Fable 5.2 Scores 41.3% on RLI, Nearly Double Rivals, as Sheets and Dashboards Roll Out — rubenhassid · 2026-11-13
- Dukoi Turns Your Old Android Phone Into a Free App Server in 2 Minutes — Scobleizer · 2026-11-13
- OpenAI DevDay Talk by Jonathan Abrams and kagigz Now on YouTube — craigsdennis · 2026-11-13
- Microsoft's .NET team open-sources 16 benchmarked agent skills, 41% fewer tokens on Sonnet 5 — victor_explore · 2026-11-13
- MikeGC launches as open-source legal AI, taking on Harvey's paid Intake features — Scobleizer · 2026-11-13