glm-5.3-flash
Open-weight sparse MoE model for efficient coding and long-horizon agentic work
3 routes serve GLM 5.3 Flash, in the order requests are offered to them. Each row is measured on its own route over the selected window, so a row with little traffic behind it reads as thin evidence rather than as the model's number.
| Provider | Context | Max output | Input/1M | Output/1M | p50 TTFT | p90 TTFT | p50 latency | p90 latency | p50 output rate | Eligible calls | Errors | Uptime | Evidence |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
Fireworks AI primary accounts/fireworks/models/glm-5p3-flash Vision | 1.04M | 131,072 | $0.15 | $0.50 | -- | -- | -- | -- | -- | -- | 0% | warming · 30m | |
Together AI failover 1 zai-org/GLM-5.3-Flash Vision | 1.04M | 131,072 | $0.15 | $0.50 | -- | -- | -- | -- | -- | -- | 0% | warming · 30m | |
NEAR AI failover 2 z-ai/glm-5.3-flash Vision | 1.04M | 131,072 | $0.075 | $0.25 | -- | -- | -- | -- | -- | -- | 0% | warming · 30m |
TTFT is the median time to the first visible answer token over real ModelRelay requests, with a fixed prompt standing in for models that have no traffic yet. Speed ranks by it. Intelligence is the Artificial Analysis Intelligence Index.
Routing follows the priority order above and fails over on error. To require a single provider, add a provider object to the request body on the OpenAI- and Anthropic-compatible endpoints. Setting allow_fallbacks to false makes the constraint strict: the request fails rather than routing elsewhere.
{"provider": {"only": ["fireworks"]}}{"provider": {"only": ["together"]}}{"provider": {"only": ["near-ai"]}}