NVIDIA's 30B MoE Model Lands on Perplexity Agent API
denisyarats · x · 2026-08-12
Perplexity announced that its Agent API now supports NVIDIA's Nemotron 3.5 Lightning model.
This is an open-source 30B MoE model with 3B active parameters, specifically built for high-volume execution layers in always-on agents, such as tool calls, validation, and subagent work. The official recommendation is to pair it with frontier models like Nemotron Ultra for planning, letting Lightning handle the volume.
Related event: NVIDIA Open-Sources Nemotron Model and Agent Routing Tools(47 posts)→
More from coding & agent
- AI Tool Usage Shouldn't Drain Mental Energy; Automation is the Goal — sujingshen · 2026-08-12
- Gated Hindsight Distillation Enhances Mobile GUI Agent Training — Weiwei Li · 2026-08-12
- Quotio: macOS Menu Bar App to Manage Multiple AI Accounts & Quotas — tom_doerr · 2026-08-12
- Principal Engineer's Guide to AI System Design Interviews: Core Bottlenecks & Fundamentals — datawithsuman · 2026-08-12
- Cursor Drops Grokbot: A Native Agent Tool with Cloud Computers & Cross-App Collaboration — victor_explore · 2026-08-12
- Qwen 27B Beats 35B in Coding on 32GB GPUs, But Local Models Still Can't Self-Check — WSTangoDelta · 2026-08-12