ECHO Paper Lands NeurIPS Spotlight: Terminal Agents Learn World Models for Free, Doubling GRPO Pass@1
DimitrisPapail · x · 2026-09-25
ECHO: Terminal Agents Learn World Models for Free
The paper "ECHO: Terminal Agents Learn World Models for Free" (Vaishnavi Shrivastava, Dimitris Papailiopoulos et al.) has been accepted as a NeurIPS 2026 Spotlight.
- Insight: CLI agents are the closest thing LLMs have to an embodied setting—the terminal's response stream (stdout, errors, logs) is a supervision signal that standard GRPO-style RL throws away, leaving failed rollouts with almost no gradient signal.
- Method: ECHO adds an auxiliary loss on top of the policy-gradient loss, training the policy to predict environment observation tokens resulting from its own actions. It reuses the same forward pass as GRPO, needs no extra rollouts, and turns terminal feedback into dense supervision.
- Results: Pass@1 on TerminalBench-2.0 roughly doubles—Qwen3-8B improves from 2.70% to 5.17%, and Qwen3-14B from 5.17% to 10.79%. ECHO policies also predict terminal dynamics far better on held-out rollouts, and from base Qwen3-8B it matches expert-SFT-then-GRPO performance without any expert demonstrations.
- Paper and code are both available.
More from coding & agent
- Pigment Agents now reads and posts in Slack, inheriting user permissions — mattturck · 2026-11-12
- Map of the agent economy: 50+ startups across payments, spend control, identity, memory — venturetwins · 2026-11-12
- OpenCode launches Browse, an on-device browser agent with local logins — Saul_Loveman · 2026-11-12
- VentureTwins maps the B2A economy: dozens of agent infra and app companies in first series installment — venturetwins · 2026-11-12
- Google releases 76-page 'Agents Companion' whitepaper on agent engineering — mdancho84 · 2026-11-12
- Spinning up containers for 20-line agent tools is backwards: Safescript compiles provable, 50ms-deploy apps — uriwa · 2026-11-12