Occamy-1.0: Open Pareto-Frontier 35B Intelligence for Co-work
Authors: Wenhui Chen, Shiwen Cheng, Hao Dong, Chenda Duan, Ruixiang Feng, Zhong Guan, Boqiang Guo, Xueyuan Han, Haojie Hao, Liangmeng Huang, Zhelong Huang, Xinke Kong, Hongyu Li, Jiazheng Li, Junbo Li, Qingchuan Li, Yukun Lian, Chang Liu, Tianyu Liu, Zicheng Liu, Shuyi Ouyang, Yijun Pan, Kunyu Shi, Xiaojun Tang, Bingquan Wang, Kesu Wang, Yuchen Wang, Sibo Wei, Sicong Xie, Xiaoying Xing, Yi Xu, Zhijun Xu, Hongwei Xue, Qingcheng Zeng, Di Zhang, Guannan Zhang, Haochen Zhang, ...
Submitted: 4 September 2026
arXiv ID: arXiv:2609.11977 (cs.AI)
Abstract
Occamy-1.0 introduces a 35-billion-parameter open-source language model engineered to sit on the Pareto frontier of performance and efficiency for collaborative AI workflows—referred to as co-work. As AI-assisted teamwork becomes standard in 2026, Occamy-1.0 addresses the need for a model that balances strong reasoning and generation capabilities with practical deployment costs. The model is fully open, enabling researchers and developers to integrate it into multi-agent systems, shared workspaces, and human-AI co-creation pipelines without restrictive licensing.
Key Contributions
- Open Pareto-frontier performance: Occamy-1.0 achieves state-of-the-art results among 35B-class models on collaborative benchmarks, outperforming several larger closed models in co-work tasks.
- Optimized for co-work: The architecture and training objective emphasize turn-taking, context sharing, and tool use—critical for seamless human-AI collaboration in 2026’s distributed work environments.
- Efficient 35B scale: By leveraging advanced sparsity and training techniques, the model delivers high throughput and low latency, making it suitable for edge and cloud deployments alike.
- Fully open: Model weights, training code, and evaluation harness are released to foster community-driven improvements.
Why It Matters in 2026
With the rise of AI agents that assist in software development, scientific research, and creative projects, the demand for models that can act as reliable co-workers has surged. Occamy-1.0 fills a gap between massive proprietary models and smaller, less capable open alternatives, offering a balanced solution that prioritizes collaboration without sacrificing performance. Its Pareto-frontier positioning means it delivers the best trade-off between accuracy, speed, and cost for real-world co-work scenarios.
Availability
Occamy-1.0 is available on arXiv and will be accompanied by open-source releases on GitHub and Hugging Face. For more details, refer to the full paper.
This article summarizes the arXiv preprint “Occamy-1.0: Open Pareto-frontier 35B Intelligence for Co-work” (arXiv:2609.11977), submitted on 4 September 2026.
via ArXiv AI