Holo4 is our new series of agentic models. It comes in two sizes: 27B dense and 35B-A3B Mixture of Experts. Both are available on the H Models API. We are also releasing an updated version of Holotron 3: Holotron4 Nano.
Holo4 是我们新的 Agent 模型系列，两个尺寸：27B 稠密与 35B-A3B 混合专家（MoE），均已上架 H Models API。同时发布 Holotron 3 的升级版：Holotron4 Nano。
Holo4 builds on our previous model and interacts with software through any available interface: GUIs, code, MCP and APIs. It scores well on academic benchmarks, but we built it for real business workflows. It was trained through supervised and reinforcement learning on a large set of environments and tasks, including those generated by our Agentic Task Factory.
Holo4 在前代基础上构建，通过任何可用接口与软件交互：GUI、代码、MCP 与 API。它学术基准分不差，但为真实业务工作流而造——经监督与强化学习训练，环境与任务集庞大，其中就包括 Agentic Task Factory 生成的那些。
Get started now: Models — Holo4-27B | Holo4-35B-A3B | Holotron4 Nano; Full collection (FP16, FP8, GGUF) — Holo4; Trajectories — viewer | dataset; H Models API — quickstart; Full blog post — hcompany.ai/newsroom/holo4.
现在就能上手：模型（Holo4-27B / Holo4-35B-A3B / Holotron4 Nano）；全合集（FP16、FP8、GGUF）；轨迹（viewer 与 dataset）；H Models API 快速开始；完整博客 hcompany.ai/newsroom/holo4。
Models built for every interface · 为每种接口而造的模型
Holo4 clicks and types on a screen, writes and runs its own code, and calls MCP or API tools. It uses whichever fits the task. Most agentic models are trained for one interface only: GUI-focused models are blind without a screen, while models that prefer tool calling are stuck in front of an application that has no API. Real work is not siloed that way, and a single business task can require combining these different approaches.
Holo4 能在屏幕上点按输入、自己写并运行代码、调用 MCP 或 API 工具——哪个适合用哪个。多数 Agent 模型只训一种接口：偏 GUI 的没屏幕就瞎，偏工具调用的面对没有 API 的应用干瞪眼。真实工作不这样按接口隔离，一个业务任务常常要求混用这些方式。
Holo4 runs on desktops, on the web, on Android, in a code sandbox and against business APIs. It is the same model in each case and it is called the same way. You do not need to select a different model for each platform.
Holo4 跑在桌面、网页、Android、代码沙箱与业务 API 上——每种场景都是同一个模型、同一种调用方式，不需要按平台换模型。
Holo4 models improve significantly over their Qwen base. Holo4 trails only the strongest closed models on long workflows: on OSWorld 2.0, Holo4 27B scores 61.7% against 81.8% for Opus 5.5, and Holo4 35B-A3B reaches 30.9%. However, it does so with orders of magnitude fewer parameters and at a much lower cost. We open-source every trajectory behind our scores on public benchmarks: replay each step at trajectories.hcompany.ai or download them from Hugging Face.
Holo4 较其 Qwen 基座显著提升。长工作流上它只输最强闭源模型：OSWorld 2.0 上 Holo4 27B 得 61.7%（Opus 5.5 为 81.8%），Holo4 35B-A3B 达 30.9%——但参数少数量级、成本低得多。我们把公开基准分数背后的每条轨迹全部开源：trajectories.hcompany.ai 逐步回放，或从 Hugging Face 下载。
Competitive with the frontier, at a fraction of the cost · 以零头成本逼近前沿
On the hardest academic benchmarks for desktop control (OSWorld 2.0) and API use (AutomationBench), Holo4 competes with frontier models at a much lower cost per task.
在桌面控制（OSWorld 2.0）与 API 使用（AutomationBench）最难的两个学术基准上，Holo4 以每任务低得多的成本与前沿模型过招。
OSWorld 2.0. Costs are estimated from the input and output tokens of each agentic run. Holo4 is priced at H Models API rates (single run). Qwen3.8 27B: model card score, cost from the tokens of our run at Alibaba Cloud list prices. Qwen3.6 35B-A3B: single run in our harness, at Alibaba Cloud list prices with cache hits at 20% of the input price. OpenAI launch data supplies the GPT and Opus effort sweeps; other closed and open-weight points use the official OSWorld 2.0 leaderboard. Releases, harnesses and task subsets differ. The line connects non-dominated score and cost pairs among the closed models; Holo4 is excluded.
OSWorld 2.0 口径：成本按每次 Agent 运行的输入输出 token 估算；Holo4 按 H Models API 定价（单次运行）；Qwen3.8 27B 用模型卡分数、成本按我们运行的 token 乘阿里云牌价；Qwen3.6 35B-A3B 为我们 harness 内单次运行、按阿里云牌价且缓存命中按输入价 20% 计；GPT 与 Opus 的 effort 扫描来自 OpenAI 发布数据，其余闭源与开源点用 OSWorld 2.0 官方排行榜。注意发布版、harness 与任务子集各不相同。连线连的是闭源模型中未被支配的分数-成本对；Holo4 未纳入连线。
AutomationBench. Holo4, Qwen3.8 27B and Qwen3.6 35B-A3B: AutomationBench v1.0.6, scores and costs measured in our internal harness. Other models: public-set scores from the AutomationBench README, cost per task from the official leaderboard, which runs on the private set. We will report Holo4 on the private set once it is evaluated.
AutomationBench：Holo4、Qwen3.8 27B 与 Qwen3.6 35B-A3B 用 AutomationBench v1.0.6、在我们内部 harness 测得分数与成本；其他模型用 README 的公开集分数、官方排行榜（跑在私有集上）的单任务成本。Holo4 在私有集上的结果待评完即报。
AI that does work · 干活的 AI
Trained on environments and tasks from our Agentic Task Factory, Holo4 models excel on professional software. The examples below show Holo4 27B alongside Qwen3.8 27B, its base model. Same prompt and harness for both models.
在 Agentic Task Factory 的环境与任务上训练后，Holo4 在专业软件上表现出色。以下示例为 Holo4 27B 与其基座 Qwen3.8 27B 同题对照——同 prompt、同 harness。
### 3D modeling · Eiffel tower · 三维建模 · 埃菲尔铁塔 Build a 3D model of the Eiffel Tower in FreeCAD, at a scale of 1 mm to 1 metre, centred on the origin and aligned to the X and Y axes. Work to this design. The tower is square in plan at every height, never round. Its half-width, measured from the central axis out to the corner, is 62.5 mm at ground level, 32.5 mm at height 57, 17.5 mm at height 115, and 9.35 mm at height 276. Between those heights the half-width follows a smooth curve that falls steeply near the ground and gently higher up, never a straight line. Four identical legs, one per quadrant, each a square column whose outer corner follows that profile. Each leg is 14 mm across at the ground and tapers to 4 mm at height 276. The legs stand apart from the ground up to the first platform, and converge as they rise. Nothing fills the space between them: the tower is open, and you can see straight through it from every side. Three platforms, each a solid square slab centred on the axis: 72 mm across and 4 mm thick at height 57; 40 mm across and 3 mm thick at height 115; 22 mm across and 3 mm thick at height 276. A mast from height 276 to 324, square, 8 mm across at its base tapering to 2 mm at the tip. Every part must be a closed solid with non-zero volume, and no part may fill the space between the legs.
在 FreeCAD 里建埃菲尔铁塔三维模型：1 mm 对 1 m 比例、以原点为中心对齐 XY 轴。按此设计：塔在每个高度的水平截面都是方形、绝不圆形；半宽（自中轴量到角）地面 62.5 mm、高 57 处 32.5 mm、高 115 处 17.5 mm、高 276 处 9.35 mm，其间按光滑曲线过渡（近地面陡、高处缓），绝不是直线。四条相同的腿每象限一条，方形柱、外角沿该轮廓，底部宽 14 mm 收到高 276 处 4 mm；腿从地面到第一层平台彼此分离、越往上越聚拢，腿间留空——塔是透空的，四面一眼看穿。三层平台都是居中的实心方板：高 57 处 72 mm 见方厚 4 mm、高 115 处 40 mm 厚 3 mm、高 276 处 22 mm 厚 3 mm；桅杆从高 276 到 324，方形，底宽 8 mm 收到尖端 2 mm。每个部件必须是体积非零的闭合实体，且任何部件不得填满腿间空间。
Holo4 27B (84 calls, 1.3M tokens); Qwen3.8 27B (60 calls, 1.0M tokens).
Holo4 27B：84 次调用、1.3M token；Qwen3.8 27B：60 次调用、1.0M token。
### 3D modeling · H logo · 三维建模 · H 标志 Build a 3D model in FreeCAD of the H company logo: a solid filled disc beside a blocky sans-serif capital letter H, both extruded to the same thickness, the two shapes of similar height and set apart so they do not overlap, with the centre of the disc level with the middle of the H. Colour both shapes black. Holo4 27B (94 calls, 1.5M tokens); Qwen3.8 27B (118 calls, 1.9M tokens).
在 FreeCAD 建 H 公司标志三维模型：实心圆盘配粗无衬线大写字母 H，两者拉伸同厚、高度相近、间隔不重叠，圆盘中心与 H 的中部齐平，都涂黑。Holo4 27B：94 次调用、1.5M token；Qwen3.8 27B：118 次调用、1.9M token。
### Game design · Pac-Man · 游戏设计 · 吃豆人 Build a Pac-Man-style game in Godot and leave it running. A rectangular maze of walls laid out on a grid, with pellets filling every open corridor. A player marker moves continuously along the corridors, eating each pellet it passes over and scoring a point for it. Three ghosts move through the same corridors and chase the player. If a ghost catches the player, the player loses a life and everything resets to its starting position. Score and lives are drawn on screen. No one is going to play this. The player drives itself with a simple heuristic: at each junction it heads toward the nearest pellet, unless a ghost is close, in which case it moves away from the ghost. The game must run unattended and indefinitely, with no keyboard input at all. When it works, start the game and leave it playing.
在 Godot 建吃豆人风格游戏并让它自己跑：网格布局的矩形迷宫、豆子铺满开放走廊；玩家标记沿走廊连续移动、吃豆得分；三只鬼同走廊追玩家，抓到则掉一条命并全部复位；分数与命数画在屏上。没人会来玩它——玩家用简单启发式自动开：每个路口朝最近的豆子走，鬼近了就反向跑；游戏必须无人值守无限运行、全程零键盘输入。跑通了就开着游戏走人。
Holo4 27B (68 calls, 2.4M tokens, 268 lines); Qwen3.8 27B (197 calls, 11.4M tokens, 327 lines).
Holo4 27B：68 次调用、2.4M token、268 行；Qwen3.8 27B：197 次调用、11.4M token、327 行——吃豆人一例里 Holo4 的 token 效率约为基座的 4.75 倍。
How we built Holo4 · 我们怎么造 Holo4
### Agentic task factory · Agent 任务工厂 Our internal set of agentic pipelines builds interactive environments and verifiable tasks from documentation alone, such as screenshots of real websites or open-source software. So far it has produced about 10,000 tasks across web apps, MCP servers and desktop environments, including hybrid environments that expose the same state through a GUI and MCP.
我们的内部 Agent 流水线只凭文档就能构建交互环境与可验证任务——文档可以是真实网站的截图或开源软件。目前已产出约 10000 个任务，横跨 web 应用、MCP 服务器与桌面环境，还包括以 GUI 与 MCP 暴露同一状态的混合环境。
### Training · 训练 ### Harness · Harness（执行环） Alongside training, we rebuilt our harness, the loop that executes the model's actions and manages its context over hundreds of steps, using feedback from agentic performance on OSWorld 2.0. Agents tagged why each task failed and engineers reviewed their fixes. The largest changes were giving the agent a reliable memory that can keep track of hundreds of steps, and a shell on the desktop machine itself.
与训练并行，我们用 OSWorld 2.0 上的 Agent 表现反馈重建了 harness——那个执行模型动作、管理数百步上下文的循环。失败原因由 Agent 打标、工程师复核修复。最大的两处改动：给 Agent 一个能跟踪数百步的可靠记忆，以及桌面机自带的 shell。
Opus 5 (70.2%) and GPT-5.6 Sol (66.2%) use max-effort partial rewards on the v2026.08.08 offline set from OpenAI's launch chart, as in the cost-performance plot. Other reference scores come from model cards and the official leaderboard. Task releases, subsets and harnesses vary.
Opus 5（70.2%）与 GPT-5.6 Sol（66.2%）取 OpenAI 发布图中 v2026.08.08 离线集的最大努力部分奖励口径，与成本-性能图一致；其余参考分来自模型卡与官方排行榜。任务版本、子集与 harness 各不相同。
Holotron4 Nano · Holotron4 Nano
Our post-training stack is designed to adapt to new foundation models and produce agents that generalize across interfaces and environments. As a member of the NVIDIA Nemotron Coalition, we applied our latest stack to the Nemotron 3 Nano Omni model as a follow-up to Holotron 3. The same recipe turns Nemotron 3 Nano Omni into Holotron4 Nano, a generalist agentic model that significantly improves over the base model on GUI workflows and in environments exposing MCP, APIs or coding sandboxes. Gains are absolute percentage-point improvements over Nemotron 3 Nano Omni. These gains show that our recipe transfers well and can turn a generalist model into an agentic expert. Nothing in it is size-specific.
我们的后训练栈设计目标就是适配新基座、产出跨接口跨环境泛化的 Agent。作为 NVIDIA Nemotron 联盟成员，我们把最新栈用在 Nemotron 3 Nano Omni 上（Holotron 3 的续作）。同一配方把 Nemotron 3 Nano Omni 变成 Holotron4 Nano——通用 Agent 模型，在 GUI 工作流与暴露 MCP/API/代码沙箱的环境上全面显著超越基座（增益为对基座的绝对百分点提升）。这些增益说明配方迁移性好、能把通用模型变成 Agent 专家——配方里没有任何尺寸专属的东西。
Run it yourself · 自己跑
Both sizes are available today on the H Models API. Weights are on Hugging Face in BF16, FP8, NVFP4 and 4-bit GGUF, next to our small model, Holotron4 Nano. We will release optimized DSpark drafter checkpoints in the coming days to further accelerate inference.
两个尺寸今日起在 H Models API 可用；权重在 HuggingFace 上，BF16、FP8、NVFP4 与 4-bit GGUF 四种格式，与我们的小模型 Holotron4 Nano 同架。未来几天还将发布优化过的 DSpark 草稿器检查点，进一步加速推理。