What happened
Vercel’s changelog announces expanded provider availability for two models: Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed. Previously, Moonshot AI was the only provider serving either model through AI Gateway. The changelog does not name the newly added providers, so it establishes a broader pool of ZDR options without specifying who those providers are.
The post describes both models as having a 1-million-token context window and accepting image input. It says K3.1 Pro improves on earlier Kimi models in coding, 3D modeling, and financial analysis, and can handle long agent runs from a small fix to a full project. Highspeed is positioned for interactive work, including coding agents, when response time matters. These are product claims in the changelog, not independent benchmark results.
Requests can use the AI SDK, Chat Completions, Responses, or Anthropic Messages APIs. The model identifiers are moonshotai/kimi-k3.1-pro and moonshotai/kimi-k3.1-pro-highspeed. Vercel also describes setup for coding agents including Claude Code, Codex, and Cursor: install the latest Vercel CLI and run vercel ai-gateway coding-agents setup, then select the desired model. The source does not provide a release date for the models themselves or detailed performance figures.
The privacy control can be set on an individual request with zeroDataRetention: true in gateway provider options, or enabled for all team requests in AI Gateway dashboard settings. The gateway then routes only to providers with a ZDR agreement. If no such provider is available for a model, the request fails rather than falling back to a provider without the agreement. The changelog says AI Gateway reflects provider pricing without a markup or platform inference fee, including BYOK, but gives no model-specific prices.
What the archive adds
This is a new access and routing fact on a known model story. Earlier Field Notes covered Moonshot’s Kimi K3.1 Pro API launch and its stated cyber-data training constraint ahead of open weights (coverage). Vercel’s changelog adds a separate operational option: more ZDR providers through AI Gateway.
Trend signal
This supports agent-ready software exposes capabilities in a limited, practical sense: a gateway exposes model access and a permission-like retention requirement that teams can apply at request or team scope. The source does not show how providers implement ZDR or establish broader interoperability, so the signal is about an explicit control boundary, not proof of end-to-end data handling.
Why now
The changelog presents this as an expansion from a single serving provider to multiple ZDR providers. That matters to teams already evaluating Kimi because they can make ZDR-required requests through the gateway without being limited to Moonshot as the only qualifying route, subject to provider availability.
Connections to other events
The change follows Moonshot’s Kimi K3.1 Pro API launch covered in the archive. That earlier event made the model available through Moonshot; this announcement concerns how Vercel AI Gateway can route it under ZDR requirements, not a new model release.
Why it matters
For teams testing coding agents or long-context workflows on Kimi, provider choice and data handling are operational constraints, not just model-selection details. A ZDR requirement that can be enforced centrally or per call gives teams a concrete way to avoid silently routing a request to a provider that does not meet that condition. Fail-closed behavior is especially relevant when a workflow may include source code, customer material, or internal documents: the trade-off is an error when compliant capacity is unavailable rather than an unintended fallback.
The expansion may also reduce dependence on a single provider for this model family, although the source does not identify the providers or explain their respective capacity, latency, or regions. Teams should therefore treat the announcement as more routing flexibility, not as evidence that every request will have a compliant route or better performance. No new price or quota is stated, and AI Gateway’s no-markup statement does not establish what an individual upstream provider charges.
Limits and uncertainty
Vercel’s changelog does not name the additional ZDR providers, describe their regions or capacity, or quantify latency and uptime. It does not provide comparative evaluations supporting the stated improvements in coding, 3D modeling, or financial analysis. The source also does not detail provider-specific data handling beyond the existence of ZDR agreements, nor does it state model prices, rate limits, or the proportion of requests each provider will serve. The claim here is a change in gateway availability and controls, not a guarantee of performance or a full account of privacy practices.
Reader takeaway
Kimi K3.1 Pro and Highspeed have more ZDR-provider options on Vercel AI Gateway, and teams can require ZDR per request or across the gateway. The practical safeguard is fail-closed routing: if no qualifying provider is available, the request fails instead of going to a non-ZDR provider. Check provider availability and your upstream costs before relying on the route in production.