Vercel halves MiniMax H3 and H3 Max video model pricing until September 13

Vercel cuts MiniMax video model pricing in half

According to the Vercel blog, MiniMax's H3 and H3 Max video generation models are being offered at half their usual price on the AI Gateway service between August 30 and September 13. The promotion runs in partnership with MiniMax and applies to any request billed through the gateway, regardless of how long the clip is or which aspect ratio a developer selects.

Two models, two trade-offs

The two models sit at different points on the quality-versus-speed spectrum. H3 is the higher-resolution option: Vercel says it produces 2K video and accepts a wide range of inputs, including a plain text prompt, a single starting image, a pair of first and last frames to define a transition, or reference material in the form of images, video and audio.

H3 Max takes the opposite approach. It renders faster at 480p and 768p, and it accepts either a text prompt or a starting image. For teams that want quick iterations or cheaper drafts before committing to a full-quality render, that positioning makes it a natural companion to the flagship model.

No code changes required

One practical detail worth noting: the model identifiers, minimax/minimax-h3 and minimax/minimax-h3-max, have not changed. Any application already sending requests to those IDs will automatically receive the discounted rate for the duration of the promotion — no redeploy, no migration, no swap to a promotional model string.

Vercel's example code uses the AI SDK's experimental video generation helper, passing a prompt, a duration and a polling configuration. Because video renders take minutes rather than seconds, the SDK runs the generation as a background job and issues short status checks until the output lands, instead of holding a single long-lived request open. Vercel also points to its asynchronous generation documentation for webhook and start-and-status route patterns for apps that prefer server-side callbacks over polling.

Getting started

Developers without an existing setup can create an API key in the AI Gateway section of the Vercel dashboard, or generate a clip directly in the browser from the model playground before writing any code. Current rates for every model live on Vercel's pricing page, and the company maintains a full list of the video models available through the gateway for comparison.

Why it matters

Video generation remains one of the most expensive line items in AI application budgets, and render costs scale with both resolution and duration — exactly the axes this discount covers. A 50% cut, applied automatically across all durations and aspect ratios, effectively doubles what a team can prototype or ship for the same spend during the two-week window.

The promotion is also a signal about where Vercel sees demand. By partnering with MiniMax to subsidise access through its own gateway, Vercel is positioning itself as the default billing and routing layer for video generation, not just text models. And the fact that a pricing change can land without any code modification is a small but real argument for routing model traffic through a gateway abstraction rather than calling providers directly.

For anyone already building on these models, there is little reason not to lean into the window: heavier test runs, longer clips or higher-resolution experiments all cost half as much until September 13.