GLM-5.3 model access discounted 50% on DigitalOcean via AI Gateway until September 8

The promotion provides a cost-saving window for engineers evaluating or running GLM-5.3, but the discount is tied to a temporary model name. Using `zai/glm-5.3-promo-50` routes requests exclusively to DigitalOcean, with no fallback to other providers if DigitalOcean fails to serve the request. This creates a single point of failure during the promo period, which may not suit reliability-sensitive workloads. The discount also disappears when the offer ends, requiring code changes to revert to the standard model name or pinned provider setup.

For production systems, the standard model name (`zai/glm-5.3`) with provider pinning is the safer choice. By setting `order: ['digitalocean']` in `providerOptions.gateway`, engineers can prioritize DigitalOcean while retaining fallback to other providers. This approach avoids disruption when the promo expires and ensures continuity if DigitalOcean experiences latency or availability issues. The trade-off is higher cost during the promo period, but the flexibility may justify the expense for critical applications.

GLM-5.3’s specifications, 1M-token context and 128K-token output, make it suitable for long-form tasks like document analysis or code generation. The discounted rate applies only to inference, not training or fine-tuning, and usage appears in spend dashboards like any other request. Engineers should treat the promo as a temporary switch rather than a permanent configuration, as hardcoding the promo name will break after September 8. The model’s integration with coding agents (e.g., Claude Code, Codex) may simplify adoption for development workflows, but the promo’s constraints still apply.