AutoEnvScaling: Agents Build Their Own Post-Training Environments, Terminal-Bench 41.1 to 53.3
rbhar90 · x · 2026-10-14
AutoEnvScaling automates post-training environment and curriculum building for terminal agents: a proposer coding agent reads the reasoning agent's rollouts, then builds, runs and fixes each Harbor task and sandbox — no human in the loop. One model plays both builder and solver, with each checkpoint creating tasks that train the next. Qwen3.6-35B-A3B improved from 41.1 to 53.3 on Terminal-Bench 2.1.
Related event: Microsoft's AutoEnvScaling Lets Agents Build Their Own RL Environments(4 posts)→
More from coding & agent
- Microsoft's AutoEnvScaling lets agents build their own RL environments, lifting Qwen3 Terminal-Bench from 41.1 to 53.3 — omarsar0 · 2026-10-15
- tiden_ai verifies AI-generated code against actual requirements, not session tests — Shruti_0810 · 2026-10-15
- Meta's Muse Wonder (Max) lands at #12 in Agent Arena at 2.1x the cost of its predecessor — arena · 2026-10-15
- "If my agent can't browse your site, I won't shop there": Amazon's agent blockade sparks backlash — mimi10v3 · 2026-10-15
- LLM Tool-Call Judges Reject Correct Values Without References: The Derivation Gap — Linhai Ma · 2026-10-15
- Debate: If 90% of Your Code Moves JSON and Writes UI, AI Should Take It — gdechichi · 2026-10-15