Tev-Gym: 1,170 dense-reward RL environments push long-horizon agentic training
RulinShao · x · 2026-10-15
Researchers introduce Tev-Gym, a suite of 1,170 environments with dense progress rewards for scaling long-horizon agentic RL. Trained on it, Tev-27B matches GPT-6 Astra and Claude Opus 5 on ALE-CM (computing & math) at Pass@5, while Tev-320B tops open models on Terminal-Bench 4.0-CM at Pass@1, just 0.02 behind GPT 5.6 sol. Framed as RL-environment infrastructure on the recursive self-improvement path.
More from coding & agent
- How to disable Claude Code idle compaction and set a custom token threshold — EricBuess · 2026-10-15
- Claude Code adds idle compaction to shrink sessions before prompt cache expires — EricBuess · 2026-10-15
- JDK 28's Built-in JSON API: Fetch and Map to Records With No Libraries — therealdanvega · 2026-10-15
- Hackathon kids keep rebuilding Git — "you do not need to build your own Git" — jia_seed · 2026-10-15
- Dev builds unhinged Gulag-themed front-end for Papercliping that show-trials failing agents — jamesbrooksco · 2026-10-15
- Agent trained on community archive tweets posts autonomously, rebuilds its own persona nightly — lfschiavo · 2026-10-15