← 研踪 PaperTrailarXiv:2611.20924 · v1
Navital Gym: Rule-Graded Agent Environments for Indian Workflows
首次提交 2026-11-16 19:35
版本更新 2026-11-16 19:35
雷达观测 2026-11-17 17:11
摘要
AI agents are being deployed on Indian workflows such as payments, tax, forms, travel and customer support in Hindi and Hinglish. Public agent evaluations rarely cover these workflows, and many score with another language model. We present Navital Gym, a set of six open environments with 321 public tasks in five task families: trip planning, UPI payments and disputes, GST invoicing and returns, form filling (including a multi-document split) and customer support in Hindi, Hinglish and three experimental Indian languages. Every task is graded by named, deterministic rules that recompute the outcome from a synthetic sandbox world. No language model judges any score. We report one run per task for three models on all six environments, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 4.5, and for Gemini 3.1 Flash-Lite on two (UPI and the Support core). Opus solves 315 of 321 tasks, Sonnet 302 and Haiku 205. Trip planning and Hindi/Hinglish support separate the models most. We also describe a private held-out set that is not published, and a 7B adapter trained in the gym. Code is Apache-2.0 and task data is CC BY 4.0. The rules are a documented simplification of Indian practice, not the live systems.
作者备注:10 pages. Code and tasks: https://github.com/Navitallabs/navital-gym
观测到的版本
相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。
v1 · 2026-11-16 19:35 · 最新观测版本
Navital Gym: Rule-Graded Agent Environments for Indian Workflows
AI agents are being deployed on Indian workflows such as payments, tax, forms, travel and customer support in Hindi and Hinglish. Public agent evaluations rarely cover these workflows, and many score with another language model. We present Navital Gym, a set of six open environments with 321 public tasks in five task families: trip planning, UPI payments and disputes, GST invoicing and returns, form filling (including a multi-document split) and customer support in Hindi, Hinglish and three experimental Indian languages. Every task is graded by named, deterministic rules that recompute the outcome from a synthetic sandbox world. No language model judges any score. We report one run per task for three models on all six environments, Claude Opus 5.5, Claude Sonnet 5.5 and Claude Haiku 4.5, and for Gemini 3.1 Flash-Lite on two (UPI and the Support core). Opus solves 315 of 321 tasks, Sonnet 302 and Haiku 205. Trip planning and Hindi/Hinglish support separate the models most. We also describe a private held-out set that is not published, and a 7B adapter trained in the gym. Code is Apache-2.0 and task data is CC BY 4.0. The rules are a documented simplification of Indian practice, not the live systems.
查看此版本解读 →查看此版本原文 ↗