Ternary Bonsai 2 27B matches 95% of Qwen's IMO score, 30% faster
tensorqt · x · 2026-09-25
PrismML benchmarked its ternary Ternary Bonsai 2 27B on the 2026 IMO problems under harsh conditions: no internet, no tools, and a 131K-token reasoning budget. The model scored in the upper end of the human bronze-medal range, retained 95% of full-precision Qwen3.8 27B's (54GB) IMO score, and finished the problems in 70% of the time. Compared against Gemma 4 12B QAT (7GB) as well. The results highlight how far ternary quantization has come for heavy reasoning tasks. Code is available on GitHub.
More from Models
- Ex-OpenAI researcher launches System One Models: Jev makes typed decisions in 70-500ms — JeremyCMorgan · 2026-09-25
- OpenAI to preview GPT-6 Cyber model and first-of-its-kind security product, per Fortune — jeremyakahn · 2026-09-25
- Vision model tier list updated with Opus 5.5, GPT-6 Sol/Luna, and Grok 4.7 — ducha_aiki · 2026-09-25
- Anthropic Accused of Quietly Nerfing Models Weeks After Launch, Opus 5.5 Expected to Follow — iannuttall · 2026-09-25
- TypeSafe AI launches Jev, a 'System One' model for bounded decisions in agent runs — hardimanjames · 2026-09-25
- Qwen Flash Next IQ4_XS beats 27B FP8 on MMLU-Pro, GPQA and GSM8K in community eval — smallDeltaBigEffect · 2026-09-25