Bonsai 2 tested: 98% of Qwen 27B in 6GB, but zero working agent builds
Prompt Engineering · youtube · 2026-09-28
The Prompt Engineering YouTube channel tested PrismML's Bonsai 2, a ternary compression scheme claiming to keep 98% of the Qwen 27B model's performance in a 6GB file.
Findings:
- On short tasks (thinking off, same agent harness) it's a tie: 48.6 vs 47.1
- On six long agentic builds, full Qwen scored 35/60 while Bonsai 2 produced zero working apps, once repeating the same search 114 times
- The video explains how ternary compression works, what PrismML's own whitepaper says about agentic benchmarks, and what trace signals to watch before trusting a compressed model with agent work
Takeaway: compressed models may be fine for chat, but think twice before handing them long-horizon agent tasks.
Related event: Bonsai 2 Compresses Qwen 27B to 6GB but Falters on Long Tasks(2 posts)→
More from Models
- User Shares Coral Reef Creation Labeled 'Qwen 4 Max (xhigh)', an Unannounced Model Name — intellectronica · 2026-11-16
- Stealth Model 'silicon-2' Surfaces in Devin, Suspected to Be Anthropic's Next Model — kimmonismus · 2026-11-16
- Google hides Gemini model label in AI Mode, rolls Gemini 4 Carbon to Pro subscribers — gaganghotra_ · 2026-11-16
- Open-weight Karpo-Decision-31B tops Decision Index at 73.9, one forward pass per answer — CodeByPoonam · 2026-11-16
- AI daily: DeepSeek rate-limits chats, Qwen haggling spurs shop warnings, Doubao workspace launch — 快鲤鱼 · 2026-11-16
- Nathan Lambert calls out Ethan Mollick for 'selectively ignoring' Kimi in model comparisons — natolambert · 2026-11-16