Perplexity Adds NVIDIA Nemotron 3 Ultra for Pro and Max Subscribers.

Perplexity has quietly rolled out access to NVIDIA's Nemotron 3 Ultra model for Pro and Max subscribers across both its core search platform and Computer product. The new model, which NVIDIA positions as purpose-built for long-running agents, expands the range of AI models available to paying users.

The announcement, shared via LinkedIn, marks another step in Perplexity's strategy of rapidly integrating new models as they become available. The company has consistently added model options for subscribers, recently integrating Claude Opus 4.8 and Claude Fable 5 as an orchestrator model in Computer.

Nemotron 3 Ultra is NVIDIA's latest open model release, designed specifically for agentic workflows that require sustained reasoning over longer periods. Unlike traditional language models optimized for quick responses, agent-focused models need to maintain context across multi-step tasks, handle tool calls, and recover from errors without human intervention.

For Perplexity users, this translates to potentially more capable Computer sessions. Computer, Perplexity's autonomous research tool, already handles complex multi-step research tasks by breaking down queries, searching across sources, and synthesizing findings. An agent-optimized model like Nemotron 3 Ultra could improve how Computer manages these longer workflows.

The model joins a growing roster of options available to Perplexity subscribers, including models from Anthropic, OpenAI, and other providers. This multi-model approach lets users select the best tool for specific tasks rather than being locked into a single provider's capabilities.

Nemotron 3 Ultra is available immediately to users on Perplexity's Pro ($20/month) and Max (pricing varies) subscription tiers. The model appears in both the standard Perplexity interface and within Computer, where it can be selected alongside other available models.

The integration follows Perplexity's pattern of making new models available shortly after their public release. NVIDIA released Nemotron 3 Ultra as an open model, meaning developers can run it locally or through various cloud providers. Perplexity's hosted implementation removes the infrastructure burden for users who want to experiment with agent-capable models without managing deployment.

For context, Computer has evolved into Perplexity's most technically ambitious product. The company recently integrated Deep Research capabilities using a search-as-code architecture, and even brought Computer inside Microsoft Office apps. Adding agent-optimized models like Nemotron 3 Ultra suggests Perplexity sees long-running autonomous tasks as a core use case.

Perplexity's rapid model integration reflects broader shifts in how AI platforms compete. Rather than building a single proprietary model, companies like Perplexity are positioning themselves as platforms that aggregate the best available models and make them accessible through clean interfaces.

This approach has tradeoffs. Users gain flexibility and can choose models based on specific needs, but the proliferation of options creates decision fatigue. For small business owners and builders using Perplexity for research or content tasks, knowing whether to use Nemotron 3 Ultra versus Claude versus GPT-4 requires understanding subtle differences in model capabilities.

NVIDIA's focus on open models also matters. While companies like Anthropic and OpenAI keep their frontier models closed, NVIDIA's strategy of releasing capable open models creates more options for developers and platforms. If Nemotron 3 Ultra performs well in agent tasks, expect to see it integrated into other tools beyond Perplexity.

For now, Pro and Max subscribers can experiment with the new model across their workflows. Whether it delivers meaningfully better results for long-running tasks remains to be seen, but Perplexity's willingness to integrate new options quickly gives users room to test and compare.