Holo3：突破桌面计算使用界限
Holo3 是面向 Autonomous Enterprise 的生产级模型系列，Holo3-122B-A10B 在 OSWorld-Verified 桌面使用基准得分 78.85%，以仅 10B 活跃参数（122B 总量）实现行业 SOTA 且成本远低于 GPT‑5.4/Opus‑4.6。通过聚焦感知与决策的 Agentic Learning Flywheel 在合成企业场景训练，能执行真实工作流。Holo3-35B-A3B 权重在 Hugging Face 以 Apache-2 开源，并可通过 Inference API（含免费层）访问。
原文内容
Back to Articles
Holo3: Breaking the Computer Use Frontier
Team
Article
Published
April 1, 2026
Upvote
–
Ramzi De Coster
ramzidecoster
Follow
Hcompany
Pierre-Louis Cedoz
plcedoz38
Follow
Hcompany
We are proud to unveil
Holo3
, the latest evolution of our vision for the Autonomous Enterprise. With a score of
78.85% on the OSWorld-Verified benchmark
, Holo3-122B-A10B establishes a new state of the art for the industry on the leading desktop computer use benchmark.
Holo3 is more than a benchmark leader; it is engineered for production. Built using our agentic flywheel, it has been trained to execute real-world workflows within synthetic enterprise environments. This not only ensures that Holo3 excels in today’s business scenarios, but establishes the foundation for a future where our agents can autonomously navigate virtually any digital landscape.
Best of all, Holo3 achieves this with only 10B active parameters (122B total), so at a fraction of the cost of large-scale proprietary models, such as GPT 5.4 or Opus 4.6. All models are available through our
Inference API
. Holo3-35B-A3B weights are openly accessible on
Hugging Face
under the Apache2 license and freely accessible through our inference API under a free tier.
The Agentic Learning Flywheel
What sets Holo3 apart is its specialized training pipeline—a continuous feedback loop designed to sharpen two core agentic pillars:
perception
and
decision-making
.
Our training flywheel is about teaching our model from annotated examples how to execute specific tasks, all while developing generalist skills across a virtually infinite variety of user interfaces. Here is how we build world-class computer use models:
Synthetic Navigation Data:
using human and generated instructions, we generate scenario-specific navigation examples.
Out-of-Domain Augmentation:
we programmatically extend the scenarios and augment the data to ensure Holo3 can handle the unexpected.
Curated Reinforcement Learning:
every data sample is careful
Holo3: Breaking the Computer Use Frontier
Team
Article
Published
April 1, 2026
Upvote
–
Ramzi De Coster
ramzidecoster
Follow
Hcompany
Pierre-Louis Cedoz
plcedoz38
Follow
Hcompany
We are proud to unveil
Holo3
, the latest evolution of our vision for the Autonomous Enterprise. With a score of
78.85% on the OSWorld-Verified benchmark
, Holo3-122B-A10B establishes a new state of the art for the industry on the leading desktop computer use benchmark.
Holo3 is more than a benchmark leader; it is engineered for production. Built using our agentic flywheel, it has been trained to execute real-world workflows within synthetic enterprise environments. This not only ensures that Holo3 excels in today’s business scenarios, but establishes the foundation for a future where our agents can autonomously navigate virtually any digital landscape.
Best of all, Holo3 achieves this with only 10B active parameters (122B total), so at a fraction of the cost of large-scale proprietary models, such as GPT 5.4 or Opus 4.6. All models are available through our
Inference API
. Holo3-35B-A3B weights are openly accessible on
Hugging Face
under the Apache2 license and freely accessible through our inference API under a free tier.
The Agentic Learning Flywheel
What sets Holo3 apart is its specialized training pipeline—a continuous feedback loop designed to sharpen two core agentic pillars:
perception
and
decision-making
.
Our training flywheel is about teaching our model from annotated examples how to execute specific tasks, all while developing generalist skills across a virtually infinite variety of user interfaces. Here is how we build world-class computer use models:
Synthetic Navigation Data:
using human and generated instructions, we generate scenario-specific navigation examples.
Out-of-Domain Augmentation:
we programmatically extend the scenarios and augment the data to ensure Holo3 can handle the unexpected.
Curated Reinforcement Learning:
every data sample is careful