NVIDIA Nemotron 3 Ultra sets accuracy record among open models in LangChain test

In July 2026, NVIDIA and LangChain published the results of a joint effort to tune an agentic framework. Nemotron 3 Ultra achieved the highest accuracy of any open model in LangChain's Deep Agents test, and it also beat leading closed models on price-to-quality ratio.

What is LangChain Deep Agents?

LangChain is the world's largest and most widely used platform for orchestrating AI agents. Millions of developers build agentic applications on it, from business-process automation to complex multi-step data analysis.

Deep Agents is LangChain's specialized framework for agents that plan in depth. Unlike simple prompt chains, Deep Agents lets the model break a task into subtasks on its own, choose tools, iterate on intermediate results and adjust the plan as it goes. This is one of the most demanding modes for a language model: every intermediate error accumulates and lowers final accuracy.

NVIDIA Nemotron 3 Ultra is an open language model from the Nemotron family, which is aimed at enterprise deployment. The LangChain team did not simply integrate it into its stack but tuned the framework specifically for the model's architecture, which is what produced the record results in agentic mode.

What Nemotron 3 Ultra achieved in the benchmark

The tuning results matter for the whole open-model market:

• Accuracy: the highest among all open models in the Deep Agents harness test
• Cost: lower than leading closed competitors at comparable quality
• Throughput: more tasks completed per unit of time
• Multiple-fold advantage: the model outperforms alternatives by 10x on one of the key operational metrics

Historically, closed models such as GPT-4o, Claude and Gemini led open ones on complex agentic tasks. An agent runs dozens of iterations, calls external tools and keeps context between steps, so every intermediate failure is costly. Paired with a Deep Agents harness tuned for it, Nemotron 3 Ultra is the first open model to break through this barrier.

"NVIDIA

Nemotron 3 Ultra offers leading performance at a lower cost than top closed models."

Why is an open model better than a closed API?

An open model with closed-API-level performance offers enterprise developers several fundamental advantages. First, data does not leave the company's infrastructure. Banks, the public sector and healthcare face strict data-localization requirements, and that is where open models have the biggest practical edge over API services.

An open model also removes dependence on a third-party provider's pricing policy: there is no risk of a sudden price hike or a change in model behavior. Inference costs can be optimized for real workloads, and the model can be fine-tuned on corporate data if needed.

Open models are also more cost-effective for agents under heavy load: deploying on your own cluster lets you scale capacity linearly without growing API-call costs.

The ecosystem effect matters too: millions of developers use LangChain, which means that whole layer of agentic applications can now run on an open model without losing performance.

What this means

An open model's win in the benchmark of the largest agent platform shifts the balance of power: closed APIs are no longer a prerequisite for high-performance agentic applications. For developers already using LangChain, Nemotron 3 Ultra becomes a real alternative, with lower cost and without sacrificing accuracy.