Vercel’s AI Gateway model catalog now lists NVIDIA’s Nemotron 3 Ultra 550B A55B with a 1 million-token context window and published pricing, alongside entries for Qwen3.7 Plus and MiniMax M3. Vercel’s documentation describes the catalog as part of a broader gateway for browsing models, comparing pricing, and routing...
Vercel’s AI Gateway model catalog lists nvidia/nemotron-3-ultra-550b-a55b as a model entry released on 06/04/2026, with a 1 million-token context window, according to Vercel AI Gateway’s public models page.
The same Vercel AI Gateway listing gives the Nemotron entry published pricing of $0.37 per million input tokens and $1.08 per million output tokens. Vercel’s catalog also shows performance figures for the model entry, including 0.3 seconds of latency and 467 tokens per second.
The catalog entry appears alongside other model options, including Qwen3.7 Plus and MiniMax M3, with pricing information also shown in Vercel’s model browser.
Vercel’s documentation for “Models & Providers” says AI Gateway supports browsing available models, filtering them by provider, viewing pricing, and querying a public REST models endpoint. That positions the catalog as a discovery and comparison surface for developers choosing among hosted AI models.
Vercel’s AI Gateway product page describes the service as a centralized interface for routing text, image, and video workloads to hundreds of AI models. Vercel also says the gateway includes billing and observability features, which are intended to help teams manage usage across providers from one place.
Taken together, the Vercel AI Gateway product page and documentation indicate that the model catalog is not only a list of available models. It is also part of a broader developer workflow for selecting models, checking prices, and connecting applications to model providers through Vercel’s gateway.
For developers, published catalog fields such as context length, input pricing, output pricing, latency, and throughput can affect model selection. A 1 million-token context window, as listed by Vercel for the NVIDIA Nemotron entry, may be relevant for applications that need to process long documents or large conversation histories. Pricing per million tokens can also materially affect costs for high-volume applications.
However, the figures should be read as catalog data from Vercel rather than independent benchmark results. The supplied Vercel AI Gateway source lists the Nemotron latency and throughput figures, while Vercel’s documentation explains that the gateway provides model browsing and pricing information. The sources do not provide details about the test conditions behind the reported latency or tokens-per-second values.
Vercel’s catalog reflects a broader pattern in AI infrastructure: application platforms are increasingly presenting many third-party models through a single routing and billing layer. Vercel describes AI Gateway in those terms, saying it can route workloads across text, image, and video models while centralizing billing and observability.
For teams already deploying applications on Vercel, that integration may reduce the need to compare model providers one by one. At the same time, developers evaluating any gateway still need to check provider availability, pricing, rate limits, model behavior, and production reliability for their own use cases.
The main confirmed update from the provided sources is that Vercel’s public AI Gateway model catalog lists NVIDIA’s Nemotron 3 Ultra 550B A55B with the stated context window, pricing, latency, and throughput figures, and that Vercel documents the catalog as part of its model browsing and routing system.
The same Vercel AI Gateway listing gives the Nemotron entry published pricing of $0.37 per million input tokens and $1.08 per million output tokens.
Vercel’s catalog also shows performance figures for the model entry, including 0.3 seconds of latency and 467 tokens per second.
The catalog entry appears alongside other model options, including Qwen3.7 Plus and MiniMax M3, with pricing information also shown in Vercel’s model browser.
Continue reading
