CoPE adds one forecasting loss to GRPO for 2x sample efficiency in agent RL
youjiaxuan · x · 2026-10-17
UIUC researchers released CoPE (Co-Evolving Acting and Forecasting for Language Agents), a simple RL method for long-horizon LLM agents. Since agentic RL returns a single end-of-trajectory reward leaving most batch samples gradient-free, CoPE adds a multi-step action forecasting loss on top of GRPO, letting agents learn action chunks from successful rollouts. This yields up to 2x sample efficiency across 5 long-horizon agent benchmarks. Code is open source (MIT) on GitHub, built on AgentGym-RL's verl fork.
More from coding & agent
- 一行 PowerShell 实现 11 秒截图转表格，作者零手写代码 — dfinke · 2026-10-17
- HeyGen demos 3-step AI video pipeline: ChatGPT script, MCP avatar, auto-editing — HeyGen · 2026-10-17
- DeepSeek launches standalone Search & Fetch APIs at ~$0.70 per 1,000 successful requests, failures free — teortaxesTex · 2026-10-17
- OpenAI's new Python and Node SDKs make browser and computer use much easier — OpenAIDevs · 2026-10-17
- .NET 10 Lambda ARM64 base image ships with 170MB of leftover Yum cache — DanielLockyer · 2026-10-17
- A one-shot cron job prompt got Claude to crack ciphers and advance a 1932 number theory bound — IgorCarron · 2026-10-17