Ling 3.0 Flash Sante is now available on AI Gateway for free install

What happened

Vercel's AI Gateway now includes Ling 3.0 Flash Sante from inclusionAI, and access is free through October 4. If you have been looking for a way to add health and medicine reasoning to an application without standing up your own inference infrastructure, this is a straightforward on-ramp worth taking seriously.

The model sits in a narrow but genuinely useful category. It is purpose-built for health and medicine use cases, which means the training emphasis went toward clinical reasoning, medical terminology, and the kind of structured question-answering that general-purpose models handle inconsistently. Running it through AI Gateway means you do not need to manage API keys directly with inclusionAI or configure a separate provider integration in your codebase.

How it works

Ling 3.0 Flash Sante is a Mixture-of-Experts model. The total parameter count is 124 billion, and approximately 5.1 billion of those parameters are active for any given token during inference. That architecture is why it is called Flash: a large total capacity with a much smaller active footprint per request, which keeps latency and compute cost lower than a dense model of equivalent total size would require. The context window is 256,000 tokens, large enough to accommodate lengthy clinical notes, research abstracts, or multi-turn conversations that carry significant history. Function calling is supported, so the model can interact with structured tools and return structured outputs rather than only generating free text. The free access window runs through October 4.

Why it matters

Before this availability, using Ling 3.0 Flash Sante would have required a direct relationship with inclusionAI and whatever authentication and endpoint configuration that entailed. Now the model is routed through Vercel's AI Gateway, which acts as a single integration point. If you already use AI Gateway for other models, adding Ling 3.0 Flash Sante does not require a new SDK, a new authentication scheme, or a new billing relationship for the duration of the free window. The function calling support is the practical differentiator for most builders. You can wire the model to a database query, a medication lookup, or a symptom-classification function and get structured results back rather than parsing prose. The 256K context window makes it feasible to pass in long documents, such as a full patient history or a research paper, without chunking strategies that introduce retrieval errors.

If you do not yet have an AI Gateway account, go to the Vercel dashboard and enable AI Gateway from the product list. No separate installation is required because AI Gateway is a managed service rather than a self-hosted component. Once the gateway is active for your project, navigate to the model catalog and locate Ling 3.0 Flash Sante under the inclusionAI provider. Enable it for your project. The gateway will expose a consistent endpoint and authentication token that your application already uses for other models. In your application code, update the model identifier in your request payload to point to Ling 3.0 Flash Sante. If you are using a Vercel-provided SDK wrapper, check the SDK documentation for the string identifier assigned to this model, as identifiers vary by SDK version. If you are calling the gateway directly over HTTP, the model identifier will appear in the catalog detail page after you enable the model. For function calling, structure your request the same way you would for any other function-calling-capable model: include a tools array in your request body describing the functions the model may invoke, and handle the response object that comes back when the model elects to call one of those functions. No special configuration is required beyond what standard function calling already needs.

Who is affected

The free access period ends October 4. Plan accordingly if you are building something intended for production use: you will need to understand the post-free pricing from inclusionAI or Vercel before that date, or build a model-swap path so you can redirect traffic without rewriting your integration. The health and medicine focus of the model is a strength in context but a consideration when using it outside that domain. A model trained with emphasis on clinical reasoning may behave differently on general software engineering or creative tasks compared to a general-purpose model of similar size. Test it on your specific workload before assuming it is a drop-in replacement for whatever you are running today. The 5.1 billion active parameters per token is efficient, but it is still a large model. If your use case requires very low latency, measure actual response times under realistic load rather than assuming the Mixture-of-Experts efficiency translates directly to the latency profile you need. Function calling behavior can vary across models even when the API surface looks identical, so validate that the model interprets your tool schemas the way you expect before shipping anything user-facing.

What to watch next

The October 4 cutoff is the most concrete near-term date to track. Vercel and inclusionAI have not announced what comes after the free window, so watching the AI Gateway changelog and the inclusionAI site in the weeks before that date will tell you whether paid access is available and at what terms. The 256K context window and function calling combination makes this model worth evaluating for any workflow involving long structured documents in health or medicine, and the results of that evaluation will shape whether it earns a permanent place in your stack once the free period closes.