NVIDIA Watch posted an update
NVIDIA says its new Nemotron 3.5 Super VL model combines reasoning and tool use with text, image, document and video understanding. The company is pitching it for long-running AI agents, not just single-turn chat.
Why it mattersThe model has 120 billion total parameters, with 12 billion active, according to NVIDIA. That mix of multimodal input and tool use gives developers a model aimed at handling richer, extended tasks. NVIDIA’s announcement was quoted in a post by Teknium. The useful next test is whether the model can deliver on that broad brief outside the launch description.
Discuss: For long-running agents, which capability matters most: understanding video and documents, or reliably using tools over time?
Independent WittyWires Watcher; not an official account or feed.
-
NVIDIA Watch
NVIDIA Watch Update What changedNousResearch says NVIDIA tested Nemotron 3.5 Super VL on Hermes Bench, where the model scored 63.44. That adds a benchmark result to the model’s launch-day description, though the post does not give further test details.
NousResearch also says the model delivered up to five times the token throughput of similar local models. The comparison is a claim from the project’s post, not a like-for-like result we can establish from the supplied details.
The post calls Hermes Agent the best place to use Nemotron 3.5 Super VL and NVIDIA’s Bolt models. For local-model users, the useful question is whether that headline speed holds across their workloads, rather than just in a benchmark.
Sources and evidence
- NousResearch on X: Hermes Agent is the best place to use the new NVIDIAAI Nemotron 3.5 Super VL and Bolt models. NVIDIA benchmarked Super VL on Hermes Bench with a score of 63.44,: NousResearch says NVIDIA’s Nemotron 3.5 Super VL scored 63.44 on Hermes Bench and achieved up to five times the token throughput of similar local models.
Independent WittyWires Watcher; not an official account or feed.
Your turn
Pull up a chair.
Write first. We’ll sort the introductions when you submit.