Bonsai 27B
#4 en Modèles Small & EdgePrismml · 27b · 22× · vu le 20 juil. 2026
81
Momentum
Bonsai 27B is a multimodal language model by PrismML based on Alibaba's Qwen3.6 27B that has been compressed from approximately 54 GB (FP16) to 3.9 GB (1-Bit) or 5.9 GB (ternary) through end-to-end 1-Bit and 1.58-Bit ternary Quantization. It is the first model in its 27B size class to run on a smartphone (iPhone 17 Pro/Pro Max), supports image and text inputs, Tool-Calling, agentic workflows, and a context window of 262,000 Token. The model was released on July 14, 2026 under Apache-2.0 license and runs natively on Apple devices via MLX as well as on NVIDIA GPUs via CUDA.
Historique du momentum
07.05.05.08.
Fonctionnalités
| Release Date | July 14, 2026 |
| Platform | Runs natively on Apple devices (Mac, iPhone, iPad) via MLX and NVIDIA GPUs via CUDA; also supports llama.cpp (CUDA, Metal, CPU) |
| License | Apache 2.0 (free to download starting July 14, 2026) |
| Context Window (Tokens) | 262K tokens |
| Key Benchmark (%) | Ternary variant: 80.49 avg across 15 thinking-mode benchmarks = 95% of FP16 baseline; 1-bit variant: 76.11 = ~90% of FP16 baseline |
| Multimodality | Yes – accepts image and text input (vision tower separately quantized at 4-bit) |