Key points
- Nemotron 3.5 Super VL is now available in ElevenAgents; on ElevenLabs' own customer-support benchmark it posted leading task-completion accuracy at 70% lower cost than other leading open models.
- ElevenLabs plans to add 15 MW of NVIDIA Vera Rubin capacity on top of its existing Blackwell Ultra systems, aimed at its largest customers.
- A voice agent for Poland's Centralna e-Rejestracja calls landline-only patients about appointments, running on-premises with NVIDIA Confidential Computing.
- Poland's system handles 40 million appointments a year, with no-show rates of 10 to 20 percent.
Why it matters for marketers: Voice agent builders get a cheaper open model option and a proven on-premises path, which matters for teams facing data-residency limits or high per-call costs.
ElevenLabs and NVIDIA announced an expanded partnership on November 11, 2026, covering models, infrastructure and secure deployments for ElevenAgents, ElevenLabs' conversational agents platform. NVIDIA's Nemotron 3.5 Super VL is now available there, and ElevenLabs is adding capacity for its biggest customers.
On ElevenLabs' benchmark for realistic customer-support tasks, the companies say Nemotron reached leading end-to-end task-completion accuracy, including on financial calculations and billing analysis, at 70% lower cost than other leading open models. Reasoning can be switched off for simpler tasks, which puts the model among the three fastest on the platform. Because the models can be post-trained on enterprise data, customers can also run their own tuned versions inside ElevenAgents.
The most concrete proof point is in Poland. ElevenLabs models and NVIDIA infrastructure power a voice agent for the central appointment system within the national e-Health platform, with EasyCall supporting delivery. The agent phones patients who only have landlines to remind them of visits and help them reschedule. The deployment runs on-premises, keeping citizen data and audio inside government infrastructure.
For marketers and customer experience teams, the takeaway is that voice agents are being built for regulated, high-volume settings where data control and cost per interaction decide the outcome. Note that the benchmark results come from ElevenLabs' own testing.
Worth watching: how the added Vera Rubin capacity reaches customers with data-residency needs, and whether sovereign deployments spread beyond Poland. The robotics and in-vehicle voice work is still exploratory, with only potential applications described so far.
Sources
This brief is our own summary. Read the full reporting at the source.