NVIDIA Launches Nemotron 3.5 Lightning: Ultra-Fast Open Model for Agents

ArtificialAnlys · x · 2026-08-11

NVIDIA has released Nemotron 3.5 Lightning, a highly efficient small open-weights model. It features 31.6B total parameters with 3.6B active parameters, utilizing a hybrid Mamba-Transformer architecture to deliver performance comparable to gpt-oss-120b at a quarter of the size.

Performance & Speed: On the Artificial Analysis Intelligence Index, it achieves exceptional time-efficiency, completing tasks in 0.5 minutes. This makes it substantially faster than open-weight peers like Qwen3.6 35B (3.5 min) and gpt-oss-120b (3.4 min). However, proprietary models like Gemini 3.5 Flash-Lite and GPT-5.6 Luna still lead the overall time-efficiency frontier.

Agentic Capabilities: With a GDPval-AA v2 Elo of 824 and a Terminal-Bench v2.1 score of 24%, it is highly attractive for agentic pipelines. NVIDIA collaborated with partners like CodeRabbit and Harvey for domain-specific post-training to enhance workflow accuracy.

Related event: NVIDIA Open-Sources Nemotron Model and Agent Router(37 posts)→

Original post →

More from Models

Models channel →