Cloudflare Runs Kimi and GLM at Scale on Workers AI

Cloudflare added Kimi and GLM to Workers AI, its globally distributed inference layer. Both are Chinese-origin models: Kimi from Moonshot AI, GLM from Zhipu AI. The addition matters less as a catalog update and more as a signal about where inference infrastructure is heading.

The framing in the announcement — smaller, faster, safer — points at a deliberate architectural direction. Smaller models at this tier typically mean lower latency per token, cheaper cost per request, and simpler deployment footprints. For engineers building latency-sensitive applications, that tradeoff is often preferable to raw capability headroom you never actually use.

Running these at scale on Cloudflare's network means inference happens close to the request origin across their edge locations. That changes the calculus for applications where round-trip latency to a centralized inference endpoint is a bottleneck. A model that is 80% as capable but returns results in half the time is often the correct production choice.

For solo founders and small teams already on Cloudflare's stack, the path to adding structured LLM calls into Workers scripts gets shorter. No separate inference provider, no cross-origin request overhead, no additional vendor contract. The integration sits inside the same runtime and billing surface.

The GLM family has strong multilingual coverage including Chinese, which opens Workers AI to use cases previously underserved by English-dominant model providers. Kimi has shown competitive long-context handling relative to its size class.

The practical question is whether Cloudflare's serving layer introduces constraints — quantization levels, context window caps, rate limits — that matter for your workload. The team's release notes are the right place to check specific model configs before committing to the integration.

This continues Cloudflare's pattern of adding capable open-weight or accessible models to Workers AI to reduce the activation energy for edge-native AI applications.

Source