Bonsai 27B: A 27B-Class model that runs on a phone
Bonsai 27B: A 27B-Class model that runs on a phone
Bonsai 27B：首款可在手机上运行的 27B 级模型
Announcing Bonsai 27B: The First 27B-Class Model to Run on a Phone 发布 Bonsai 27B：首款可在手机上运行的 27B 级模型
Today, we’re announcing Bonsai 27B, based on Qwen3.6 27B, the new multimodal flagship of the Bonsai family and the first model of its capability class to run on a phone. 今天，我们正式发布 Bonsai 27B。该模型基于 Qwen3.6 27B 构建，是 Bonsai 系列全新的多模态旗舰产品，也是同等能力级别中首个能在手机上运行的模型。
Our earlier releases proved that models with 1-bit and ternary weights could produce commercially useful language models. Bonsai 27B extends that frontier to a new capability tier: multi-step reasoning, structured tool calls, vision tasks, and computer-use agentic loops that stay coherent across many steps. Until today, deploying that tier locally has been impractical for a concrete reason: a 27B model occupies roughly 54GB in 16-bit precision, and even a good 4-bit build, at 18GB, is too large for a phone and for most laptops. 我们之前的发布证明了使用 1-bit 和三进制权重的模型可以产生具有商业价值的语言模型。Bonsai 27B 将这一前沿技术扩展到了一个新的能力层级：多步推理、结构化工具调用、视觉任务以及能够在多个步骤中保持连贯的计算机使用代理循环。直到今天，在本地部署这一层级的模型仍不切实际，原因很具体：一个 27B 模型在 16-bit 精度下占用约 54GB 空间，即使是优秀的 4-bit 构建版本也需要 18GB，这对手机和大多数笔记本电脑来说都太大了。
Bonsai 27B changes that. It comes in two variants: Bonsai 27B 改变了这一点。它提供两个版本：
Ternary Bonsai 27B uses ternary {−1, 0, +1} weights with FP16 group-wise scaling, giving a true 1.71 effective bits per weight. At 5.9 GB, it is the quality-oriented variant: it runs on an everyday laptop with the full reasoning, tool-calling, and agentic capability. 三进制 Bonsai 27B 使用三进制 {−1, 0, +1} 权重和 FP16 分组缩放，实现了每个权重 1.71 个有效位的真实表现。其大小为 5.9 GB，是面向质量的版本：它可以在日常笔记本电脑上运行，并具备完整的推理、工具调用和代理能力。
1-bit Bonsai 27B uses binary {−1, +1} weights with the same group-wise scaling, giving 1.125 effective bits per weight. At 3.9 GB, it is the footprint-oriented variant, which fits within the memory budget of an iPhone 17 Pro, bringing a 27B-class model onto a phone for the first time. 1-bit Bonsai 27B 使用二进制 {−1, +1} 权重和相同的分组缩放，实现了每个权重 1.125 个有效位。其大小为 3.9 GB，是面向占用空间优化的版本，能够适配 iPhone 17 Pro 的内存预算，首次将 27B 级模型带到了手机上。
As with every Bonsai release, the low-bit representation runs end to end across the language network, embeddings, attention, MLPs, and the LM head, with no higher-precision escape hatches. Both variants are multimodal, with the vision tower shipping in a compact 4-bit form so on-device workflows can see screenshots, documents, and camera input, not just text. Bonsai 27B carries a full 262K-token context, and supports speculative-decoding, compounding the speed with lossless draft-and-verify acceleration. Everything is available today under the Apache 2.0 License. 与以往的 Bonsai 发布一样，低位表示法在整个语言网络、嵌入层、注意力机制、MLP 和语言模型头中端到端运行，没有任何高精度回退机制。两个版本均为多模态，视觉塔以紧凑的 4-bit 形式提供，因此设备端工作流不仅能处理文本，还能识别截图、文档和摄像头输入。Bonsai 27B 支持完整的 262K token 上下文，并支持投机解码，通过无损的“草稿与验证”加速进一步提升速度。所有内容即日起在 Apache 2.0 许可下可用。
Retaining the intelligence
保持智能水平
Across a 15-benchmark suite spanning knowledge, reasoning, math, coding, instruction following, tool calling, and vision (evaluated in thinking mode, where the model’s full reasoning is exercised) Ternary Bonsai 27B retains 95% of the full-precision baseline, and 1-bit Bonsai 27B retains 90%. 在涵盖知识、推理、数学、编码、指令遵循、工具调用和视觉的 15 项基准测试套件中（在“思考模式”下评估，即发挥模型的全部推理能力），三进制 Bonsai 27B 保留了全精度基准的 95%，而 1-bit Bonsai 27B 保留了 90%。
Read the table by capability and the story is sharper than the averages: math and coding are nearly untouched, tool calling stays within a few points of full precision - exactly the capabilities that agentic workloads depend on. For comparison, the most aggressive conventional low-bit build of the same base model scores significantly lower than 1-bit Bonsai 27B while occupying 2.5x more memory. 按能力查看表格，结论比平均值更清晰：数学和编码能力几乎未受影响，工具调用能力与全精度相比仅有几个百分点的差距——这正是代理工作负载所依赖的核心能力。相比之下，同一基础模型最激进的传统低位构建版本，在占用 2.5 倍内存的情况下，得分仍显著低于 1-bit Bonsai 27B。
This is the same Pareto shift we demonstrated with our earlier language and image models, now at 27B scale: 27B-class capability at a footprint smaller than a full-precision 2B model. By intelligence density — the measure we introduced with 1-bit Bonsai 8B — 1-bit Bonsai 27B delivers 0.53 per GB: more than 10x the full-precision baseline, and roughly 2.7x the best low-bit alternative available. 这是我们之前在语言和图像模型中展示的帕累托改进，现在扩展到了 27B 规模：以小于全精度 2B 模型的大小，实现了 27B 级别的能力。根据我们随 1-bit Bonsai 8B 引入的“智能密度”指标，1-bit Bonsai 27B 每 GB 提供 0.53 的智能密度：是全精度基准的 10 倍以上，约为目前最佳低位替代方案的 2.7 倍。
Why this is an important paradigm shift
为什么这是一个重要的范式转移
The most valuable AI workloads are shifting from single responses to sustained work: assistants that operate real tools, workflows that run unattended before returning a result, and research that synthesizes dozens of documents. This shift changes the shape of the workload — an agent doesn’t make one model call, it makes hundreds, each one carrying context, producing structured output, and feeding the next. 最有价值的 AI 工作负载正在从单次响应转向持续工作：能够操作真实工具的助手、在返回结果前自动运行的工作流，以及能够综合数十份文档的研究任务。这种转变改变了工作负载的形态——代理不再只进行一次模型调用，而是进行数百次调用，每一次都携带上下文、产生结构化输出，并为下一次调用提供输入。
Cloud APIs will remain the right choice for many products. But for agentic workloads, cloud-only execution imposes structural constraints: every step is a remote request, per-token cost accumulates with every iteration, and every plan, tool call, and intermediate result crosses the network including the user’s private files, screen, and data. 对于许多产品而言，云 API 仍然是正确的选择。但对于代理工作负载，仅依赖云端执行会带来结构性限制：每一步都是远程请求，每迭代一次 token 成本就会累积，且每一个计划、工具调用和中间结果都需要经过网络传输，包括用户的私有文件、屏幕和数据。
Local execution changes the equation. When a model capable of sustained agentic work fits on the device, the agent can live inside the product: the marginal cost of a hundred-step loop is zero, and the user’s data never leaves the machine. Entire categories open up — persistent on-device agents, assistants that work offline, assistants that reason over private local data by construction. What has been missing is a model small enough to deploy this way and capable enough to trust with the work. Bonsai 27B is that model. 本地执行改变了这一局面。当一个具备持续代理工作能力的模型能够装入设备时，代理就可以驻留在产品内部：百步循环的边际成本为零，且用户数据永远不会离开设备。这开启了全新的类别——持久化的设备端代理、离线工作的助手、以及能够直接对本地私有数据进行推理的助手。此前缺失的正是这样一个既足够小以实现这种部署，又足够强大以胜任工作的模型。Bonsai 27B 就是这样的模型。
It also unlocks a new system architecture: hybrid deployments that route non-frontier and privacy-sensitive tasks to a capable local model and reserve frontier cloud models for the hardest steps — collapsing the cost-per-task of agentic systems. 它还开启了一种新的系统架构：混合部署。将非前沿且对隐私敏感的任务路由到强大的本地模型，并将最困难的步骤留给前沿云模型，从而大幅降低代理系统的单任务成本。
Bonsai 27B reaches up to 163 tok/s in 1-bit and 134 tok/s in Ternary on an NVIDIA GeForce RTX 5090. On an M5 Max, it reaches up to 87 tok/s in 1-bit and 58 tok/s in Ternary. 在 NVIDIA GeForce RTX 5090 上，Bonsai 27B 的 1-bit 版本最高可达 163 tok/s，三进制版本最高可达 134 tok/s。在 M5 Max 上，1-bit 版本最高可达 87 tok/s，三进制版本最高可达 58 tok/s。
Fitting a phone is a stricter gate than storage numbers suggest. A phone never exposes its full memory to an app - a 12 GB iPhone offers about 6 GB for the model to use on-device, and the model shares that budget with its KV cache and activations. 适配手机比存储数字所暗示的门槛要严格得多。手机从不会向应用程序开放其全部内存——一部 12 GB 的 iPhone 大约只提供 6 GB 给设备端模型使用，而且模型还需要与 KV 缓存和激活值共享这一预算。