今天,我们宣布推出 Bonsai 27B,该模型基于 Qwen3.6 27B,是 Bonsai 系列的新一代多模态旗舰模型,也是首个能在手机上运行的同能力等级模型。
我们之前的发布已经证明,采用 1 比特和三进制权重的模型能够产生具有商业实用价值的语言模型。Bonsai 27B 将这一前沿拓展到了新的能力层级:多步推理、结构化工具调用、视觉任务,以及能在多个步骤中保持连贯性的计算机使用智能体循环。在此之前,在本地部署该层级模型因一个具体原因而不可行:一个 27B 模型在 16 比特精度下占用约 54GB,即使是优秀的 4 比特版本也需 18GB,对于手机和大多数笔记本电脑来说都太大了。
Bonsai 27B 改变了这一局面。它提供两个版本:
三进制 Bonsai 27B 采用三进制 {−1, 0, +1} 权重,并配合 FP16 分组缩放,实现每个权重仅 1.71 有效比特。该版本大小为 5.9 GB,是注重质量的版本:可在日常笔记本电脑上运行,具备完整的推理、工具调用和智能体能力。
1 比特 Bonsai 27B 采用二进制 {−1, +1} 权重,配合相同的分组缩放,实现每个权重仅 1.125 有效比特。该版本大小为 3.9 GB,是注重占用空间的版本,可适配 iPhone 17 Pro 的内存预算,首次将 27B 级别的模型带入手机。
与每次 Bonsai 发布一样,低比特表示贯穿语言网络、嵌入向量、注意力机制、MLP 和语言模型头的端到端运行,没有任何更高精度的逃逸通道。两个版本均为多模态,视觉塔以紧凑的 4 比特形式提供,使得设备端工作流不仅能处理文本,还能处理截图、文档和摄像头输入。Bonsai 27B 拥有完整的 262K 模型 token 上下文窗口,并支持推测解码,通过无损的草稿-验证加速机制进一步提升速度。所有内容今日起在 Apache 2.0 许可证下开放。
保持智能水平
在涵盖知识、推理、数学、编程、指令遵循、工具调用和视觉(以思考模式评估,即模型进行完整推理)的 15 项基准测试套件中,三进制 Bonsai 27B 保留了全精度基线的 95%,而 1 比特 Bonsai 27B 保留了 90%。
| 类别(基准测试) | Qwen 3.627B | 三元 Bonsai27B | 1-bit Bonsai27B |
|---|---|---|---|
| 数学(GSM8K, MATH-500, AIME25, AIME26) | 95.3 | 93.4 | 91.7 |
| 编程(HumanEval+, MBPP+, LiveCodeBench) | 88.7 | 86.0 | 81.9 |
| 智能体与工具调用(BFCL v3, TauBench) | 80.0 | 74.0 | 66.0 |
| 指令遵循(IFEval, IFBench) | 78.4 | 71.8 | 65.8 |
| 知识 / STEM(MMLU-Redux, MuSR) | 83.1 | 77.0 | 73.4 |
| 视觉(MMMU Pro, OCRBench) | 72.6 | 65.2 | 59.6 |
| 总体(15 项基准测试) | 85.0 | 80.5 | 76.1 |
图 I:Bonsai 27B(思考模式)与全精度基线的基准测试得分。完整的逐项基准测试结果见白皮书。
按能力维度阅读该表,其揭示的信息比平均值更为鲜明:数学和编程能力几乎未受影响,工具调用能力与全精度版本的差距也仅在几个百分点之内——这正是智能体工作负载所依赖的能力。作为对比,同一基础模型最激进的常规低位构建版本,其得分显著低于 1-bit Bonsai 27B,同时却占用 2.5 倍的内存。
这与我们之前在语言和图像模型上展示的帕累托优化是同一模式,现在扩展到了 27B 规模:27B 级别的能力,占用空间却小于全精度的 2B 模型。以智能密度(即我们在 1-bit Bonsai 8B 中引入的衡量标准)来衡量,1-bit Bonsai 27B 每 GB 提供 0.53 的智能密度:是全精度基线的 10 倍以上,并且大约是现有最佳低位替代方案的 2.7 倍。

为何这是一个重要的范式转变
最有价值的 AI 工作负载正从单次响应转向持续性工作:操作真实工具的智能助手、在返回结果前无人值守运行的工作流,以及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——一个智能体并非只调用一次模型,而是调用数百次,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。
对于许多产品而言,云 API 仍将是正确的选择。但对于智能体工作负载,纯云端执行会带来结构性限制:每一步都是一次远程请求,每次迭代都会累积每个 token 的成本,并且每个计划、工具调用和中间结果都需通过网络传输,其中包含用户的私人文件、屏幕和数据。
轮播图 I:基于 NVIDIA GeForce RTX 5090 上我们自研的 Ternary Bonsai 27B 模型,实现的端到端智能体工作流。
本地执行改变了游戏规则。当一个能够持续执行智能体任务的模型可以部署在设备上时,智能体就能融入产品内部:百步循环的边际成本为零,且用户数据永不离开设备。全新的应用类别由此打开——持久化的设备端智能体、可离线工作的助手、以及天然能基于本地私有数据进行推理的助手。此前一直缺少的,是一个足够小巧、能以这种方式部署,同时又足够强大、值得信赖的模型。Bonsai 27B 正是这样的模型。
它还解锁了一种新的系统架构:混合部署模式,将非前沿和隐私敏感的任务路由给能力强大的本地模型处理,而将前沿云模型保留给最困难的步骤——从而大幅降低智能体系统的单任务成本。
Bonsai 27B 在 NVIDIA GeForce RTX 5090 上,1-bit 模式下可达 163 tok/s,Ternary 模式下可达 134 tok/s。在 M5 Max 上,1-bit 模式下可达 87 tok/s,Ternary 模式下可达 58 tok/s。
适配手机是一个比存储数字所显示的更严格的限制。手机永远不会向应用暴露其全部内存——一部 12 GB 的 iPhone 大约只有 6 GB 可供模型在设备端使用,并且模型还需要与它的 KV 缓存和激活值共享这部分预算。任何传统构建方式的 27B 模型都无法接近这个门槛。而 1-bit 的 Bonsai 27B 仅约 4 GB,是首个能够通过这一限制并留有运行空间的模型。
正是这一限制,使得该模型系列特意设定了两个明确的工作模式,并始终将此考量铭记于心:Ternary 模式用于笔记本级别的质量,1-bit 模式用于手机级别的体积。
演示 II:基于 1-Bit Bonsai 27B 在 iPhone 17 Pro Max 上驱动的多模态智能体应用场景(演示模式:缓存与预填充图像上下文)
前沿在不断推进。
每一次 Bonsai 的发布都让“每 GB 智能密度”的边界向左移动,而 Bonsai 27B 更是将其推过了一个实用门槛:一个现代模型所具备的全部能力——思考能力、多模态理解、视觉能力、可靠的工具使用——如今已经可以装进人们已有的设备中。
我们相信,智能密度将成为 AI 下一阶段发展的关键衡量维度之一。原始能力决定了模型能做什么;密度则决定了它能在哪里做。每一次边界的左移,都扩大了先进 AI 可以运行的设备、产品和环境范围,并改变了它所触及的每一个部署场景的经济性——从手机到单 GPU 服务。Bonsai 背后的方法论与架构无关,而这条边界还将继续移动:更大的模型和新的架构已经在推进中。
早期的计算机占据整个房间;如今它们装在我们的口袋里。智能正在经历同样的旅程,而 Bonsai 27B 是迄今为止迈出的最大一步。
Bonsai 27B 可通过 MLX 在苹果设备(Mac、iPhone、iPad)上原生运行,也可通过 CUDA 在 NVIDIA GPU 上运行,这得益于为其混合注意力架构构建的定制低位内核。模型权重现已根据 Apache 2.0 许可证开放。随着本次发布,我们还提供限时免费的开发者预览 API,方便开发者轻松试用我们的模型。
关于压缩、评估和基准测试流程的完整技术细节,请参阅我们的白皮书。
加入我们
PrismML 源自加州理工学院的一个研究团队,在 Khosla Ventures、Cerberus 和 Google 的支持下创立,并持续获得三星的支持。我们花了数年时间攻克该领域最棘手的问题之一:在不牺牲神经网络推理能力的前提下对其进行压缩。
如果你想帮助构建下一代最先进的 AI,我们期待你的加入。请查看我们的招聘页面。
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.
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.
Bonsai 27B changes that. It comes in two variants:
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.
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.
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.
Retaining the intelligence
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%.
| Category (benchmarks) | Qwen 3.6 27B | Ternary Bonsai 27B | 1-bit Bonsai 27B |
|---|---|---|---|
| Math (GSM8K, MATH-500, AIME25, AIME26) | 95.3 | 93.4 | 91.7 |
| Coding (HumanEval+, MBPP+, LiveCodeBench) | 88.7 | 86.0 | 81.9 |
| Agentic and Tool-calling (BFCL v3, TauBench) | 80.0 | 74.0 | 66.0 |
| Instruction following (IFEval, IFBench) | 78.4 | 71.8 | 65.8 |
| Knowledge / STEM (MMLU-Redux, MuSR) | 83.1 | 77.0 | 73.4 |
| Vision (MMMU Pro, OCRBench) | 72.6 | 65.2 | 59.6 |
| Overall (15 benchmarks) | 85.0 | 80.5 | 76.1 |
Fig I: Benchmark scores of Bonsai 27B (thinking mode) against the full-precision baseline. Full per-benchmark results are in the whitepaper.
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.
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.

Why this is an important paradigm shift
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.
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.
Carousel I: End-to-end agentic workflow with Hermes, powered by our Ternary Bonsai 27B model on NVIDIA GeForce RTX 5090.
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.
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.
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.
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. No conventional build of a 27B model comes close to clearing it. At about 4 GB, 1-bit Bonsai 27B is the first to pass through with room to work.
That constraint is why the family ships two deliberate operating points, specifically keeping that in mind: ternary for laptop-class quality, 1-bit for phone-class footprint.
Demo II: Multimodal agentic use-cases powered by 1-Bit Bonsai 27B on an iPhone 17 Pro Max (Demo Mode: Cached & Prefilled Image Context)
The frontier keeps moving
Every Bonsai release has moved the intelligence-per-gigabyte frontier left, and Bonsai 27B moves it past a practical threshold: the full capability set of a modern model with thinking, multimodal understanding, vision, reliable tool use, now fits on the devices people already own.
We believe intelligence density will be one of the defining axes of the next stage of AI progress. Raw capability determines what a model can do; density determines where it can do it. Every leftward shift of the frontier expands the set of devices, products, and environments where advanced AI can operate and changes the economics of every deployment surface it touches, from phones to single-GPU serving. The methodology behind Bonsai is architecture-agnostic, and the frontier will keep moving: larger models and new architectures are already in progress.
Early computers filled rooms; today they live in our pockets. Intelligence is making the same journey, and Bonsai 27B is its largest step yet.
Platform Coverage
Bonsai 27B runs natively on Apple devices (Mac, iPhone, iPad) via MLX and on NVIDIA GPUs via CUDA, through custom low-bit kernels built for its hybrid-attention architecture. Model weights are available today under the Apache 2.0 License. With this release, we’re offering a free, limited-time developer preview API so developers can easily try our model.
Full technical details of our compression, evaluation, and benchmarking processes are available in our whitepaper.
Join Us
PrismML emerged from a team of Caltech researchers and was founded with support from Khosla Ventures, Cerberus, and Google, with continuing support from Samsung. We've spent years tackling one of the field's hardest problems: compressing neural networks without sacrificing their reasoning ability.
If you want to help build the next generation of state-of-the-art AI, we'd love to hear from you. Check out our careers page.