we_are_coded.by CODE · The world, decoded
БГ

Vercel / Moonshot · models and access

Kimi K3 by Moonshot now runs through Vercel AI Gateway

Vercel added Moonshot AI's new model to its AI Gateway - the one door through which an app calls many models. For developers that means one extra line of code, not a new integration. Access starts with the identifier moonshotai/kimi-k3.

Source
Vercel Changelog
Published
The event
Checked on
Responsible editor
Cvetelin Ivanov
How we work
MethodCorrections

Vercel's AI Gateway is one door in front of many models: you write to it, and it routes the request to the chosen provider, without you changing your code for every new model. Today Kimi K3 joins that list - the newest model from China's Moonshot AI. The change for the developer amounts to dropping in the name moonshotai/kimi-k3.

Fact

The facts: On 16 July 2026 Vercel announced that Moonshot AI's Kimi K3 is available through AI Gateway. AI SDK identifier: moonshotai/kimi-k3. Per Vercel and Moonshot: a 1-million-token context window, built-in reading of images and video besides text, and a reasoning mode that stays on permanently. The goal, per them, is long development tasks and working with code alongside visual context (frontend, games, CAD). On pricing, Vercel claims it reflects the provider's price with no markup and no platform fee on requests. Source: Vercel Changelog, vercel.com/changelog/kimi-k3-is-now-available-on-ai-gateway (16.07.2026).

Let's wait and see if it holds up. A new model appeared, they put it on the gateway in a day - that's the easy part, that's how this pipe works and exactly why it's convenient. The hard part is whether the model does the job live, not in the promo. Vercel describes the strengths (code plus visual, long context), but there are no benchmarks in the announcement, and the claims themselves - permanent reasoning, video reading - are Moonshot's and Vercel's words, not something we've measured. The good part here is the price: it's the provider's, with Vercel not adding anything on top. That counts.

Here's what you're actually buying: not a new mind, but a one-line difference. Today you try GLM, tomorrow Kimi K3, the day after whatever comes out - without touching your architecture. That freedom to swap models in seconds is worth more than any single model, because it keeps you from being tied to one provider. For us it slots in right alongside the free ones - a real task with code and an image attached will show whether it's worth the noise around it. For now the pipe wins, the model still has to prove itself.

The visual is generated code-art, no third-party images.