Breaking

LangChain Integrates NVIDIA Nemotron Models Into Agent Stack

June 29, 2026 at 17:03 EDT

AI Agents
Foundation Models
Open Source

NVIDIA said LangChain has integrated with its open LLM family Nemotron, providing a production-ready stack that spans entire agent workflows from inference to orchestration. Built around customization, tuning, and self-deployment, the setup lets developers run frontier-grade agent performance on their own terms.

Mar 16, 2026 · LangChain × NVIDIA

The Nemotron stack goes production: open agents, end-to-end

LangChain integrates NVIDIA's open Nemotron model family across the full agent workflow — inference to orchestration — served as OpenAI-compatible NIM APIs and callable through ChatNVIDIA.

5.9×

Ultra throughput vs GLM-5.1

3.3×

Nano throughput vs Qwen3-30B-A3B on one H200

token context, NVFP4 quantization

−30%

cost on Ultra long-running agents

Nemotron 3 Ultra throughput, drawn to scale

Relative inference throughput on long-running agent workloads (NVIDIA benchmarks).

5.9×

vs GLM-5.1

1.6×

vs Qwen-3.5

1×

baseline

The model family

Nemotron 3 Nano

30B-A3B class

3.3× throughput vs Qwen3-30B-A3B on a single H200.

Nemotron 3 Super

120B total / ~12B active

Core model for production agents.

Nemotron 3 Ultra

550B total / 55B active · MoE

5.9× vs GLM-5.1, up to 30% lower cost. Day 0 LangChain support.

In the field · Greptile code review

Ultra in beta cut real costs

4.5×

cheaper reply classification

−62%

cheaper indexing

The full enterprise stack

LangGraph · LangSmith · Deep Agents → ChatNVIDIA → NIM (OpenAI-compatible) → Nemotron + NeMo Agent Toolkit

What works

OpenAI-compatible — easy migration of existing code
Strong cost & speed advantages for agent building
Open weights, data and recipes under NVIDIA Open Model License

Watch out for

Higher latency observed on free-tier testing

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.