Summary: On February 13, 2026, Cloudflare confirmed that Zhipu’s GLM-4.7-Flash is live on Workers AI, alongside agents-on-Cloudflare tooling and Workers AI Provider v3.1.1. The pairing gives Zhipu a low-friction distribution channel and gives Cloudflare another model in its serverless AI catalog. The article notes that edge deployment trades direct control of model versions and selection for convenience.
February 13, 2026: a community post from Cloudflare confirmed that GLM-4.7-Flash, built by Zhipu, is now live on Workers AI. The same post also delivered the agents-on-Cloudflare tooling and pushed Workers AI Provider to version 3.1.1.
The date lands in the middle of Zhipu’s own milestone week — GLM-5 had gone open source on February 11. Two days on, a Zhipu model was running on Cloudflare’s serverless platform.
One Side Builds the Model, the Other Runs It
What makes the pairing work is specialization. Zhipu builds the models; Cloudflare brings the distribution: Workers AI hosts inference inside a serverless environment, so a developer can invoke GLM-4.7-Flash without provisioning a single GPU or signing an enterprise contract.
For Zhipu, the arrangement opens a low-friction channel into markets beyond its home market. For Cloudflare, each model added to the catalog pushes the platform closer to being the default runtime for AI applications. A Flash-tier model belongs in exactly this spot — workloads that watch both cost and latency are the ones edge deployment rewards most.
The ergonomics of deployment count for as much as the geography. Workers AI runs alongside everything else in a Cloudflare application — routing, storage, cron, KV — so an inference call becomes one more function call inside a stack the team already operates, not a new vendor relationship to manage. For small teams, collapsing procurement and architecture into a single step is often what separates shipping an agent feature from postponing it.
Agent Tooling Rides Along in the Same Post
There is more here than a catalog gaining one entry. Agents-on-Cloudflare tooling and Workers AI Provider v3.1.1 showing up side by side signals a single ambition: Cloudflare is after more than inference — it wants the environment agents execute in, with model, tools, and runtime folded into one developer surface.
Place the post next to February’s headline launches — agent teams in Opus 4.6, long-horizon tasks in GLM-5 — and the shape becomes clear: agent capability is compounding at both ends at once. The models keep getting stronger; the platforms keep getting simpler to assemble. What a developer finally experiences as an agent stack is the sum of those two.
The versioned provider update is the quiet giveaway. Workers AI Provider v3.1.1 is plumbing — the library that wires models into the platform’s developer surface — and bumping its version in step with a model launch shows Cloudflare treating AI model access as maintained infrastructure with release discipline of its own, not a catalog bolted onto a CDN.
What You Trade Away in Practice
Handing model execution to a third-party edge platform swaps control for convenience:
- The version and availability calendar belongs to the platform, which means release notes now demand tracking on two fronts
- Low latency and zero ops come with the edge placement, but model selection is bounded by whatever the catalog carries
- When volume is high and costs matter, a Flash-tier model on serverless billing is typically the combination worth piloting first
- Self-hosting still wins wherever data residency or custom fine-tunes cannot be compromised; the edge catalog trades that control for speed to market
None of these trade-offs is permanent — but choosing deliberately which ones to accept is what separates an architecture from an accident.