GameVerse dataset enables long-horizon interactive world modeling
GameVerse contains 69.7K clips totaling 1,655 hours with stable camera trajectories for training extended interactive models.
Stories per day · 30 days
4 RSS
Relatedagents4qwen3reinforcement-learning3ai-agents2autonomous-agents2gemini2minecraft2mineflayer2multi-agent2agent-evaluation1agent-harness1agentic-ai1alibaba1benchmarks1coding-agents1context-management1continual-learning1dataset1
49 stories, newest first · one card per story
Stories tagged #long-horizon
HaMo combines experience cards with continual learning for web agents
The framework beats harness-only and model-only adaptation on within-domain and cross-domain workflow shifts.
RiseHarness iteratively improves terminal-agent harnesses with weights fixed
It studies failed versus successful runs of the same task and adapts shared interaction mechanisms, not model weights.
Sage-A1.5 is a 35B MoE model for long-horizon agents
It adds subagent delegation, progress checks with strategy revision, and exploratory research beyond predefined questions.
Taste-Bench finds top models weak at long-term agent choices
GPT-5.6 Sol scored 59.7 percent on 502 binary forks; trained Qwen3.6-27B advice more than doubled SWE-bench Pro success.
12
TreeOPSD adds turn-level credit for long-horizon agent RL
The critic-free method raised Qwen3-30B-A3B-Thinking-2507 accuracy by 4.97 points over GRPO on Deep Research.
A team of Gemini 4 Argon agents beats Minecraft
Four Gemini 4 Argon agents autonomously defeated Minecraft's Ender Dragon in 27 hours using only Mineflayer.
A team of Gemini 4 Argon agents beats Minecraft
Four Gemini 4 Argon agents autonomously defeated Minecraft's Ender Dragon in 27 hours using Mineflayer.
Kolab benchmarks human-agent collaboration on long-horizon workflows
The dataset covers 115-plus hours from 50 professionals in 40 occupations; frontier models overestimate collaboration quality.
MMSearch-OPD: Easy Made Hard, Hard Made Learnable for Long-Horizon Multimodal Search Agents
MMSearch-OPD hardens multimodal questions and jointly trains long-horizon search agents with RL and distillation.
Interlock synthesizes completion graphs to train long-horizon agents
Fine-tuning two Qwen3.6 backbones on accepted trajectories lifts AutomationBench pass rates by 4.00-4.50 points.
CarryOn organizes long-horizon agent context around task objects
Authors report gains of 5.8% and 7.0% with token cuts of 51.9% and 22.9% on two benchmarks.
Alibaba Qwen Introduces QwenGyre: Elastic Reinforcement Learning RL for XLong-Horizon Agents
Alibaba Qwen introduced QwenGyre, elastic RL that speeds extreme long-horizon agent training.
Recursive Environment Scaling for General Long-Horizon Agents
Recursive Environment Scaling hardens agent benchmarks and improves fine-tuning by up to 14.61 points.
EvoWeb benchmark tests coding agents on evolving web apps
It uses 320 cumulative requests across 32 applications and a live-browser tester for integration and regressions.
ProgAgent: Learning Tool Orchestration through Programmatic Training
ProgAgent trains LLMs to orchestrate tools via code, beating direct tool-calling baselines including at 64k-token states.
ProgramGym: Scaling Training Environments for Long-Horizon Program Reproduction Agents
ProgramGym supplies 1,648 verifiable environments and raises Qwen3.8-27B program-reproduction accuracy from 34.87% to 42.84%.
Scaling Agentic Data for Long-Horizon Terminal Intelligence
Researchers fine-tune Qwen3.5-27B into TerminalHorizon-27B using 1.5K long-horizon terminal environments.
FlowState keeps execution state as memory for long-horizon agents
Against a full-context DeepSeek-V4-Flash baseline, authors report higher pass rates and about 40% fewer tokens.
FoldBack: Self-Correcting Masked Generative Policy for Long-Horizon Garment Folding
FoldBack lets a folding robot undo and retry failed grasps without retraining.
SGF+: Decoupling Gradient Flows for Autoregressive Video Generation
SGF+ decouples video context writing from denoising, enabling 24-hour generation from 5-second training rollouts.
SkillForge: Co-Evolving Skills and Agents via Dynamic Skill Lifecycles
SkillForge co-evolves agent skills through a fitness lifecycle, lifting success rates while retiring obsolete skills.
Does an Agent's History Tell You When Compaction Will Hurt? A Modest, Bounded Effect on the TRACE Paired-Replay Corpus
On TRACE's 590 AppWorld boundaries, recent agent history only weakly predicts harmful context compaction.
Test-Time Agent Evolution for Long-Horizon Legal Reasoning
A training-free legal agent adapts at test time using memory evolution and rubric-aligned collaboration.