InclusionAI's 560B-parameter Ling 3.1 Flash lands free on Vercel AI Gateway

Ling 3.1 Flash arrives on AI Gateway

Vercel has added Ling 3.1 Flash, a hybrid reasoning model from InclusionAI, to AI Gateway, its unified API for calling language models. According to the Vercel changelog, the model is free to use through October 13, 2026.

The headline numbers are substantial. Vercel lists Ling 3.1 Flash at 560 billion total parameters, with 25 billion active per token — a layout characteristic of sparse mixture-of-experts designs, where only a fraction of the network fires for any given token and inference stays comparatively cheap. On AI Gateway, the model is served with a 262K-token context window.

Vercel positions the model toward software development, multi-step analysis and agents that call tools, including workflows built around lengthy documents, code and long-running task histories.

Two model IDs, two end-of-promotion behaviors

Developers can call the model under two names: inclusionai/ling-3.1-flash and inclusionai/ling-3.1-flash-free. Both are free while the promotion runs, but they diverge once it ends. The standard identifier switches over to paid billing, while the -free identifier simply stops serving requests.

That distinction is worth settling before anything reaches production. An application wired to the standard model ID will keep working after October 13 and start accruing charges; one wired to the -free variant will fail instead. For evaluation and prototypes, either name works. For a deployment where unexpected costs are worse than downtime, the -free ID is the safer default — and the reverse holds if silent failure is the bigger risk.

Vercel also points developers to its AI Gateway coding agents guide, which covers selecting the standard model ID for agent setups, and to the Gateway's model playground for quick experimentation alongside the wider model catalog.

What the Gateway contributes

AI Gateway is Vercel's single interface for model calls: one API for invoking models, tracking usage and cost, and configuring routing, retries and failover. Authentication is flexible — developers can use an AI Gateway API key or bring their own provider key, so the new model can slot into existing setups without a separate credential flow.

Why it matters

Free access to a 560B-parameter reasoning model is the obvious draw. It lowers the cost of benchmarking a large hybrid reasoning model against the incumbents on your own workloads, with a context window measured in hundreds of thousands of tokens to stress-test along the way. The positioning toward agents and multi-step tool use matches where much of the industry's attention currently sits, and a long context window is a practical requirement for the extended task histories those workloads generate.

The promotion mechanics also deserve attention. A free window with divergent end-of-promotion behavior is a sensible pattern, but it pushes an operational decision — pay or fail — onto developers up front. Teams trying the model should decide which failure mode they prefer now, rather than discovering it on October 13.

For Vercel, the addition feeds the broader Gateway pitch: a growing catalog of models behind one API, where swapping Ling 3.1 Flash in and out is a model-name change rather than an integration project.