Kimi K3.1 Pro gains more Zero Data Retention providers on Vercel AI Gateway

More Zero Data Retention options for Kimi models

Vercel has expanded Zero Data Retention (ZDR) coverage for Moonshot AI's Kimi K3.1 Pro and its faster variant, Kimi K3.1 Pro Highspeed, across its AI Gateway. According to the Vercel blog, Moonshot AI had been the only provider serving either model on the platform until this change, leaving ZDR-constrained customers with a single routing option. Requests for both models can now be handled by a wider pool of providers that hold ZDR agreements.

What the models do

Per Vercel, K3.1 Pro builds on earlier Kimi releases with stronger performance in coding, 3D modeling and financial analysis, and it is built to sustain long agent sessions, from a small fix all the way up to a full project. Both variants offer a context window of one million tokens and accept image input. The Highspeed edition targets interactive work such as coding agents, where turnaround time on each response matters.

How ZDR enforcement works

Two mechanisms control the requirement. Setting zeroDataRetention in the gateway provider options makes ZDR mandatory for an individual request, while a toggle in the AI Gateway dashboard extends the requirement to every request a team sends, including traffic originating from coding agents. Once either is active, the gateway routes only to providers operating under a ZDR agreement. If no qualifying provider exists for a given model, the request fails outright rather than quietly falling back to one without such an agreement — a fail-closed design that treats the privacy constraint as non-negotiable.

Calling the models

Developers can reach both models through the AI SDK, the Chat Completions, Responses and Anthropic Messages APIs, or any coding agent connected to the gateway. The identifiers are moonshotai/kimi-k3.1-pro for the standard version and moonshotai/kimi-k3.1-pro-highspeed for the faster one.

For tools such as Claude Code, Codex and Cursor, Vercel advises upgrading to the latest Vercel CLI and running vercel ai-gateway coding-agents setup, then selecting one of the two Kimi models inside the agent. The models can also be tried in the model playground.

Gateway background

AI Gateway serves as a unified API layer for invoking models, tracking usage and cost, and configuring retries, failover and performance optimizations that Vercel says deliver higher uptime than individual providers alone. It ships with custom reporting, ZDR support, budgets for API keys and routing rules. Vercel passes provider pricing through without markup and levies no platform fee on inference, including on Bring Your Own Key requests.

Why it matters

Zero data retention agreements are often a hard requirement for teams in regulated industries, because they determine whether prompts and outputs are stored by the underlying provider. Until now, those teams had exactly one place to send Kimi K3.1 Pro traffic on the gateway, which meant a single provider outage or capacity problem could cut off access entirely. Adding more ZDR providers restores the failover and redundancy that make a gateway attractive in the first place, without weakening the retention guarantee.

The fail-closed routing behavior matters just as much: a request that cannot satisfy the ZDR constraint fails visibly instead of silently reaching a provider that might log the data. For organizations balancing agentic coding workflows against data-handling obligations, this closes a practical gap between capability and compliance.