Cloudflare AI Search adds GLM-5.3 Flash with a 1M-token context

Cloudflare has added Z.ai's GLM-5.3 Flash to AI Search for text generation. The August 30 changelog entry gives Cloudflare customers another way to use the model's 1,048,576-token context window: inside a managed search workflow running on Workers AI, rather than only through a direct model endpoint.

What Cloudflare changed in AI Search

AI Search instances can now select @cf/zai-org/glm-5.3-flash as their text-generation model. Cloudflare's entry confirms the model runs on Workers AI and points users to the supported-model configuration for setup. The update is an availability change, not a new GLM release: Z.ai released GLM-5.3 Flash on August 26, and Cloudflare added the model to Workers AI that day.

That distinction matters for teams tracking deployment options. The new entry connects the model to retrieval-augmented applications that already use Cloudflare's indexing and search layer. It can reduce the number of separate provider integrations in a stack, although the changelog does not promise identical behavior across all AI Search configurations.

GLM-5.3 Flash brings long context and multimodal input

Z.ai's model card describes GLM-5.3 Flash as the first natively multimodal model in the GLM-5 series, with 320 billion total parameters and 18 billion active parameters. It accepts text, images, and video as input and is designed for coding and long-horizon agent tasks. The model card also documents a hybrid sparse-and-linear attention architecture aimed at lowering long-context serving cost.

OpenRouter independently lists the model at a 1M-token context size, with text output and multiple providers, including Cloudflare. Its current model page shows standard pricing of $0.05 per million input tokens and $0.1667 per million output tokens on the listed route, while provider pricing and promotions vary. Those figures are not Cloudflare's AI Search price; users should check Cloudflare's own Workers AI pricing for the deployment they choose.

What developers should verify before switching

The useful test is whether a specific AI Search application can use the model's context capacity effectively. Check the supported-model configuration, retrieval limits, output behavior, rate limits, and billing for the chosen Workers AI plan. A 1M-token model context does not mean every indexed corpus or single request will receive 1M tokens.

Teams handling images or video should also confirm how their AI Search pipeline passes non-text inputs. Z.ai documents multimodal model capability, while Cloudflare's new entry specifically describes the AI Search use as text generation. That wording leaves the supported input path inside AI Search narrower than the model's full capability until Cloudflare documents otherwise.

For implementation patterns around retrieval, tools, and agent workflows, see LinkLoot's AI workflow automation guide.

Sources and scope

Cloudflare is the primary source for the August 30 AI Search availability change. OpenRouter independently corroborates the public model listing, context size, provider availability, and current displayed pricing. Z.ai's public model documentation was consulted for architecture and multimodal details, but is not included as a second source here because its model-card URL was already used in earlier LinkLoot coverage. This article covers the documented availability change and does not claim that every Cloudflare account or AI Search configuration has identical access.