NVIDIA Open-Sources Nemotron-Terminal: Boosts Qwen3-32B from 3.4% to 27.4% on Terminal-Bench 2.0

_weiping · x · 2026-09-25

NVIDIA researcher Renjie Pi announced that the team's paper has been accepted to the NeurIPS 2026 Datasets and Benchmarks Track. Nemotron-Terminal is a systematic data engineering pipeline for scaling LLM terminal agents, bridging open and proprietary models via a fully open synthetic-to-real trajectory pipeline.

Key result: SFT on Nemotron-Terminal-Corpus lifts Qwen3-32B from 3.4% to 27.4% on Terminal-Bench 2.0 (+24.0 points), rivaling much larger models.

Components:

Terminal-Task-Gen: a lightweight curation pipeline combining adaptation of existing datasets with robust synthetic task construction;
Nemotron-Terminal-Corpus: a massive open-source dataset of diverse terminal interactions with explicit planning and execution trajectories.

Original post →

NVIDIA Open-Sources Nemotron-Terminal: Boosts Qwen3-32B from 3.4% to 27.4% on Terminal-Bench 2.0

More from coding & agent