At a glance
- Super VL and Bolt 74B are available; Bolt E4B is scheduled for October 23.
- Super VL targets visual documents and multi-step agent tasks.
- Model memory figures do not cover every runtime overhead.
What is available
NVIDIA announced Nemotron 3.5 Super VL and Bolt models on October 15. Super VL and Bolt 74B-A9B 24GB are available according to the release. Bolt E4B 4GB is scheduled for October 23 and was not yet released when this article was checked. Source 1
The capability focus
Super VL combines visual understanding, document information and tool use for agent workflows. Bolt targets local and edge deployment, with low-bit quantization in the 74B-A9B 24GB version. Performance and ranking claims in the announcement are vendor-supplied; we have not independently reproduced them.
Our analysis: map the entire workflow
For enterprise documents, decide what the model reads, where outputs go and which actions need approval. Open weights create deployment choices but do not solve data quality, permissions or exception handling. Test a document through its complete review and handoff process.
Account for runtime overhead
A memory figure in a model name is not an assurance that an entire application fits on a device. Context caches, inference software, the operating system and other applications also consume resources. Measure startup time, long-document performance and concurrent use.
A practical starting point
Read the original model card and license, confirm framework and hardware support, then run a small test with non-sensitive material. A public model release does not establish that the model is available through Youjie’s API channels.
Sources & verification
Primary materials consulted for this article. Vendor measurements are not independent Youjie tests.
- NVIDIA expands Nemotron with Super VL and Bolt models ↗NVIDIA · Source date 2026-10-15 · Checked Oct 16, 2026