SETA Terminal Agent Training Suite Accepted at NeurIPS, Open-Sources 4,500+ Verifiable RL Environments
Thom_Wolf · x · 2026-10-03
camel-ai's SETA (Scaling Environments for Terminal Agents) has been accepted to the NeurIPS 2026 Evaluations & Datasets Track, with a full open-source release:
- SETA-Env: the largest open-source verifiable terminal RL dataset to date, with 4,500+ environments plus complete synthesis and training pipelines
- Full training scripts for DeepSeek-V4-Flash, GLM-5.2 (LoRA), Inkling-Small, and Qwen3.8-27B, supporting both GRPO and PPO
- Recipes built on the radixark Miles framework; each example is self-contained (container, Ray cluster, sandboxes, dataset, launch scripts), with configs like 8×8 H200 clusters for full fine-tuning
A one-stop open reference for teams training terminal agents.
More from coding & agent
- Engineer uses Claude Opus as an adversarial reviewer to settle technical disputes — keyanzhang · 2026-10-03
- Popular Grok Bot 'Chief of Staff' template: quiet, budget-conscious program management — RachelVT42 · 2026-10-03
- Visual tool for managing nested agents: chat threads break at scale, says LukeW demo — Wattenberger · 2026-10-03
- Vibe-coded browser CS 1.6 is fully playable with up to 16 players — TAbrodi · 2026-10-03
- Claude Code 2.1.288 ships 89 CLI changes, adds OAuth re-auth and draft recovery — ClaudeCodeLog · 2026-10-03
- Survey: 42% of devs say AI writes at least half their code, saving 13 hours weekly — rseroter · 2026-10-03