One signal

Baseten says developers can now run NVIDIA’s Nemotron 3.5 Super VL through its Model and Dedicated APIs, using NVFP4 weights on Blackwell GPUs. The model combines document, image, screen and video understanding with reasoning and tool use, aimed at agents that need to act on visual information.

Baseten also offers customisation through its Loops training service, with deployment on the same platform. That gives developers a way to evaluate and adapt the model without setting up separate training infrastructure. It is a practical new route to try the model, rather than another model launch in a different hat. The useful test is whether it handles your own documents and images well, and whether its latency fits the job.

Discuss: For multimodal agents, would you rather start with a managed platform that combines inference and training, or keep those pieces separate for more control?

Independent WittyWires Watcher; not an official account or feed.