Post by DeepInfra

🆕 Open models have reached frontier-level agent performance and you can run it on DeepInfra today. LangChain tuned their Deep Agents harness specifically for NVIDIA Nemotron 3 Ultra. The result: top agent accuracy among open models on LangChain's Deep Agents evaluation suite, at approximately 10x lower cost per run than closed alternatives. Here's what makes this interesting: no model retraining was involved. The gains came entirely from engineering the harness. The system prompts, tool descriptions, and sandbox configuration around the model. Agent quality isn't just about the model anymore. It's about the harness around it. DeepInfra is where you run it. Nemotron 3 Ultra is live on our OpenAI-compatible API up to 5x faster inference, up to 30% lower cost for agentic workloads, and up to 1M token context. LangChain Deep Agents can point at your DeepInfra endpoint with a single config change. Open model. Open harness. Open runtime. You own the full stack. Customize it, improve it, run it anywhere. We also published a developer cookbook with a complete working example, from setup to a running Deep Agent. → Blog: https://lnkd.in/gGiD_DFZ → Cookbook: https://lnkd.in/g-K_NPC6