Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards

Hugging Face Daily Papers Papers
long-horizon-agents reinforcement-learning reward-shaping agentic-environments training-dataset

Summary

本文提出Tev-Gym——首个具有可验证、密集渐进奖励的长周期智能体训练数据集（7个领域1170个任务），并基于进度-势能奖励塑形（PPRS）训练出Tev-27B，在Agents' Last Exam计算与数学科学领域匹配GPT-6 Astra和Claude Opus 5的Pass@5表现，并可泛化至Terminal-Bench 3.0/4.0。

Long-horizon agents have drawn growing attention for their ability to work autonomously -- from hours to days -- on economically valuable tasks, a capability essential to automated discovery and recursive self-improvement (RSI). Yet open-source models still struggle on long-horizon tasks, largely for lack of both training data and a working recipe, and scaling both is hard: long, complex tasks are difficult to curate and verify, and their rewards are sparse. We address both. First, we introduce Tev-Gym, the first long-horizon agent training dataset with verifiable, dense, progressive rewards -- 1,170 verified task environments across 7 domains. It is built on a latent-world-first approach that first constructs a latent world: a complete and correct hidden specification of the task domain, workflow, and expected state after each step. Each task is then progressively rendered from this specification into an agent-facing environment, enabling controllable complexity and a natural progress metric. Second, we turn this metric into a training recipe: via progress-potential reward shaping (PPRS), we convert progress into a dense per-step reward that improves training efficiency and makes otherwise-intractable tasks trainable. Training with this recipe yields Tev-27B, which achieves substantial gains on the Agents' Last Exam computing and math sciences domains (ALE-CM), matching GPT-6 Astra and Claude Opus 5 at Pass@5. Despite training only on ALE-related data, it generalizes to the Terminal-Bench (TB) 3.0 and 4.0 subsets, surpassing GLM 5.2 and Gemini 3.7 Flash, and reaches Claude Opus 4.8-level performance on middle-horizon tasks such as TB 2.1.
Original Article
View Cached Full Text

Cached at: 10/14/26, 06:50 AM

Paper page - Tev: Scaling Long-Horizon Agentic Environments with Dense Progress Rewards

Source: https://huggingface.co/papers/2610.18868

Abstract

Long-horizonagentshavedrawngrowingattentionfortheirabilitytoworkautonomously--fromhourstodays--oneconomicallyvaluabletasks,acapabilityessentialtoautomateddiscoveryandrecursiveself-improvement(RSI).Yetopen-sourcemodelsstillstruggleonlong-horizontasks,largelyforlackofbothtrainingdataandaworkingrecipe,andscalingbothishard:long,complextasksaredifficulttocurateandverify,andtheirrewardsaresparse.Weaddressboth.First,weintroduceTev-Gym,thefirstlong-horizonagenttrainingdatasetwithverifiable,dense,progressiverewards--1,170verifiedtaskenvironmentsacross7domains.Itisbuiltonalatent-world-firstapproachthatfirstconstructsalatentworld:acompleteandcorrecthiddenspecificationofthetaskdomain,workflow,andexpectedstateaftereachstep.Eachtaskisthenprogressivelyrenderedfromthisspecificationintoanagent-facingenvironment,enablingcontrollablecomplexityandanaturalprogressmetric.Second,weturnthismetricintoatrainingrecipe:viaprogress-potentialrewardshaping(PPRS),weconvertprogressintoadenseper-steprewardthatimprovestrainingefficiencyandmakesotherwise-intractabletaskstrainable.TrainingwiththisrecipeyieldsTev-27B,whichachievessubstantialgainsontheAgents’LastExamcomputingandmathsciencesdomains(ALE-CM),matchingGPT-6AstraandClaudeOpus5atPass@5.DespitetrainingonlyonALE-relateddata,itgeneralizestotheTerminal-Bench(TB)3.0and4.0subsets,surpassingGLM5.2andGemini3.7Flash,andreachesClaudeOpus4.8-levelperformanceonmiddle-horizontaskssuchasTB2.1.

View arXiv pageView PDFProject pageAdd to collection

Get this paper in your agent:

hf papers read 2610\.18868

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper2

#### gordonhu/Tev-320B Text Generation• 321B• Updatedabout 1 hour ago • 2 • 8 #### gordonhu/Tev-27B Text Generation• 27B• Updatedabout 1 hour ago • 2 • 8

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2610.18868 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2610.18868 in a Space README.md to link it from this page.

Collections including this paper0

No Collection including this paper

Add this paper to acollectionto link it from this page.

Similar Articles

Cowork-Gym: Synthesizing Interactive Work Environments at Scale

Hugging Face Daily Papers

Cowork-Gym 通过 build-execute-revise 循环与有状态用户模拟器大规模合成协作式 agent 环境（64.9K 个、14 个应用场景），并用于后训练 9B/27B/35B-A3B Qwen 模型，在六个 agentic 基准上分别提升 8.7、6.4、4.1 分，其中 27B 版本在 PawBench 和 ClawEval 上取得最高平均分。

DeltaTTG: Training Multi-Turn Agents with a Token-to-Go Progress Meter

Hugging Face Daily Papers

DeltaTTG 提出一种可复用的进度计量器（progress meter），以 Token-to-Go（剩余生成 token 数）作为完成进度的可观测代理，在不改变奖励的前提下重新分配多轮智能体训练中的优势值，在 ALFWorld、PaperSearchQA 和 BFCL v3 上显著优于 GRPO 等 RL 基线。