Z.ai's GLM 5.3 Flash arrives on Vercel AI Gateway with 1M-token multimodal context

What's new

Vercel has added GLM 5.3 Flash, a model from Z.ai, to the catalog of its AI Gateway service, according to a changelog post published on August 26. The company positions it as a lower-cost, quicker variant of GLM 5.3, tuned for coding and agent workloads that unfold over many steps rather than single-shot prompts.

Model capabilities

According to Vercel, GLM 5.3 Flash accepts both text and image input, which makes it suitable for tasks where the model has to reason about a screenshot, a design mockup or other visual material alongside written instructions. Its context window spans 1 million tokens, which matters for agent loops where tool results, retrieved documents and conversation history accumulate quickly. On the API side, the model supports function calling, structured output and streaming responses.

The changelog demonstrates a representative use case: passing in a dashboard screenshot and asking the model to reproduce it as a responsive React component. Images can be supplied either as plain URLs or as Base64 data URLs, and a single request can carry several images at once. The model is addressed through the gateway with the identifier zai/glm-5.3-flash.

Wiring it into coding agents

Vercel also highlights how the model fits into its coding agent setup. Running vercel ai-gateway coding-agents setup connects tools such as Claude Code, Codex, OpenCode, Cursor and Pi to the gateway, after which developers can select zai/glm-5.3-flash as the backing model inside their agent of choice. For quick evaluation, the model is also available in Vercel's model playground.

What the gateway adds

The gateway itself is pitched as a unified API in front of multiple model providers. Beyond routing requests, it handles usage and cost tracking, configurable retries, failover between providers, and performance optimizations that Vercel claims deliver higher uptime than the underlying providers alone. Additional features listed in the changelog include built-in custom reporting, Zero Data Retention support, per-API-key budgets and routing rules.

On pricing, Vercel says the gateway mirrors provider pricing with no markup and charges no platform fee on inference, including for requests made under a Bring Your Own Key arrangement.

Why it matters

Coding agents are unusually sensitive to both cost and latency, because a single task can involve dozens or hundreds of model calls. A faster and cheaper model with a 1M token context window is a good match for that profile: long context keeps multi-step sessions coherent, while lower per-call cost and latency compound across an entire run.

Multimodal input is the other half of the equation. Frontend and UI coding tools routinely need to work from screenshots rather than written specs, and vision support means agents built on this model can act on visual artifacts directly instead of relying on a separate vision step.

Finally, distribution matters. Being reachable through AI Gateway means GLM 5.3 Flash sits behind the same API as the other models Vercel offers, with no added inference fee, so teams can try it against their current model with minimal friction and switch back if results or costs disappoint.