Salesforce Koa：面向智能体工具使用的企业级语言模型
Salesforce Koa: An Enterprise Language Model for Agentic Tool Use
Salesforce 推出企业级语言模型 Koa，基于开源 Nemotron-3-Super-120B 用 GRPO 强化学习后训练，并以 FP8 部署，专攻智能体工具调用。在 CRMAgentBench 上 Koa 任务成功率达 87%，高于 GPT-4.1 的 82% 和基座的 79%，并在内部生产工具调用基准上保持领先，同时未损失 Tau2Bench、BFCL 等通用能力。
Authors:Zixiang Chen, Sufeng Niu, Yingchi Liu, Wenting Zhao, Akshara Prabhakar, Shubham Mehrotra, Bin Bi, Zhujun Lan, Katherine Tan, Mohammad Ramezanali, Tulika Manoj Awalgaonkar, Monojit Banerjee, Jielin Qiu, Shiva Kumar Pentyala, Zhepeng Cen, Anupam Tripathi, Ali Ziaei, Regunathan Radhakrishnan, Darvish Lee Shadravan, Shelby Heinecke, Sitaram Asur, Jayesh Govindarajan, Silvio Savarese, James Zhu, Phil Mui, Huan Wang
Abstract:We present Salesforce Koa, an enterprise language model built by post-training the open-weight Nemotron-3-Super-120B foundation model with reinforcement learning using Group Relative Policy Optimization (GRPO), and deployed in FP8 for production. Koa is trained only on public and synthetically generated data, and specialized for the agentic tool use that enterprise workflows demand: routing a request to the correct action, invoking the right tool with valid arguments, and completing multi-turn business tasks. The distinctive component of our pipeline is specification-driven task construction: declarative Agent Script specifications are expanded into persona-conditioned multi-turn environments whose rewards are grounded in successful tool use. Applied to enterprise CRM specifications, the same pipeline produces the in-domain training distribution on which Koa is specialized. On CRMAgentBench and the human-labeled production tool-calling set, Koa outperforms both its untuned open-weight base and GPT-4.1 and is competitive with the strongest frontier models. It reaches 87% Task Success Rate on CRMAgentBench (vs. GPT-4.1 at 82% and the base at 79%), is at or near the top of every metric on the human-labeled portion of an internal production benchmark, and preserves the base model's general capability on public benchmarks (Tau2Bench, BFCL). A controlled comparison with architecture and RL recipe held fixed shows that the additional in-domain RL stage improves argument accuracy and full tool-call success on the human-labeled enterprise benchmark.
| Comments: | 17 pages, 1 figure, 11 tables |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI); Machine Learning (cs.LG) |
| Cite as: | arXiv:2609.15066 [cs.CL] |
| (or arXiv:2609.15066v2 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2609.15066 arXiv-issued DOI via DataCite |
Submission history
From: Zixiang Chen [view email]
[v1]
Mon, 14 Sep 2026 05:30:28 UTC (3,974 KB)
[v2]
Thu, 8 Oct 2026 17:16:07 UTC (65 KB)
来源：arXiv cs.AI · arxiv.org