Mistral Large 4 comes to Vercel AI Gateway with single API key access

Mistral Large 4 arrives on AI Gateway

Vercel has added Mistral Large 4 to AI Gateway, its unified endpoint for calling AI models. According to the Vercel blog, the model is open-weight and built to handle text and images natively, with reasoning folded in. Vercel suggests it fits workflows in cyber defense, manufacturing and finance.

The detail that matters most to developers is access: the model can be reached with a single API key, and no separate Mistral account or signup is needed. Teams already sending model traffic through the gateway can fold Mistral Large 4 into their existing setup instead of provisioning new provider credentials.

How to call the model

On the gateway, the model is served under the identifier mistral/mistral-large-4. Vercel says it can be used with the AI SDK, an OpenAI-compatible Chat Completions endpoint, the Responses API, or the Anthropic Messages API, so code written against any of those interfaces can reach it with little adjustment.

Coding agents connected to AI Gateway can select the model too. The documented path is to run npx vercel ai-gateway setup and pick mistral/mistral-large-4 in the agent's model settings. Developers who want to evaluate the model before wiring it into an application can try it in Vercel's model playground or browse the wider model catalog.

What the gateway layer does

According to the announcement, AI Gateway gives developers a single API for invoking models, keeping tabs on usage and spend, and setting up routing, retries and failover. Authentication is flexible: either an AI Gateway API key or a bring-your-own provider key works. In practice, that centralises model switching, cost monitoring and resilience policies in one place rather than spreading them across separate provider integrations.

Why it matters

The release lowers the barrier to trying Mistral's newest large model. One key and no extra account means teams already on Vercel's tooling can experiment with a multimodal reasoning model, or shift production traffic to it, as a configuration change rather than an integration project.

The open-weight angle is notable as well. Models with downloadable weights appeal to organisations that want the option to self-host or inspect what they are running, and gateway access offers a managed way to use such a model before committing to dedicated infrastructure of their own.

Support for OpenAI-compatible, Responses and Anthropic Messages interfaces also means integrations are not tied to one vendor's SDK shape, which keeps migration costs down as the model landscape keeps shifting.