Observed facts: Vercel’s changelog says Nemotron 3.5 Lightning is available at no charge on AI Gateway under nvidia/nemotron-3.5-lightning-free, served by NVIDIA. Observed facts: The same source lists a 1M-token context window, tool calling, structured outputs, and warns that the provider may retain prompts for model improvement.
Mechanism: The gateway makes the model selectable through familiar APIs and connected coding agents, so an existing integration can test it without adopting a separate serving path. The price change reduces the cost of that trial; it does not remove the need to inspect the provider’s data terms or validate outputs on representative work.
Why now: A free route makes experimentation easier now, while the explicit retention note makes data governance part of the decision rather than an afterthought.
First implication: Teams can run a bounded evaluation through their existing gateway, using non-sensitive inputs until they have checked applicable retention terms and measured task quality.
Evidence boundary: The supplied Vercel changelog establishes availability, interface options, listed capabilities, and a prompt-retention caveat. It provides no independent benchmark, usage limits, service-level guarantee, or evidence that access is free outside AI Gateway.
First watch condition: Watch for published usage limits or changes to prompt-retention terms, and compare task quality on a controlled workload before routing production traffic.
This is a new analytical record because no prior New Runtime record was supplied for this specific gateway access and pricing change.