Z.ai: GLM 5.2 Performance
Try GLM 5.2 on us. Starter credit covers your first requests, then one key reaches all 32 provider routes with automatic failover.
z-ai/glm-5.2
Measured performance
Continuously sampled p50/p95 time-to-first-token (TTFT), effective throughput, and success rate for Z.ai: GLM 5.2. Effective throughput uses provider-reported output tokens over complete request time. Unsupported route and probe-configuration rows are separated from provider downtime, and no prompt or output content is stored.
| Provider | p50 TTFT | p95 TTFT | Effective throughput | Uptime | Config excluded | Availability samples |
|---|---|---|---|---|---|---|
| friendli | 586 ms | 1055 ms | 188 tok/s n=1 | 100.00% | — | 4 |
| parasail | 898 ms | 3616 ms | 204 tok/s n=2 | 100.00% | — | 2 |
| digitalocean | 961 ms | 1200 ms | 65 tok/s n=1 | 100.00% | — | 2 |
| boundless | 1026 ms | 1026 ms | 166 tok/s n=2 | 100.00% | — | 1 |
| databricks | 1179 ms | 1179 ms | 251 tok/s n=1 | 100.00% | — | 1 |
| together | 1185 ms | 1749 ms | 211 tok/s n=2 | 100.00% | — | 3 |
| scaleway | 1459 ms | 1459 ms | 110 tok/s n=2 | 100.00% | — | 1 |
| phala | 1504 ms | 14656 ms | 23 tok/s n=1 | 66.67% | — | 6 |
| byteplus | 1526 ms | 2426 ms | 45 tok/s n=1 | 100.00% | — | 2 |
| io-net | 1564 ms | 2759 ms | 73 tok/s n=1 | 100.00% | — | 2 |
| morph | 1770 ms | 1820 ms | 86 tok/s n=1 | 100.00% | — | 3 |
| lyceum | 1890 ms | 23446 ms | 152 tok/s n=2 | 100.00% | — | 4 |
| zai | 1923 ms | 2241 ms | 57 tok/s n=2 | 100.00% | — | 2 |
| regolo | 2044 ms | 4078 ms | 52 tok/s n=1 | 100.00% | — | 2 |
| tencent | 2217 ms | 4118 ms | 34 tok/s n=2 | 100.00% | — | 4 |
| novita | 2237 ms | 2237 ms | 87 tok/s n=2 | 100.00% | — | 1 |
| cloudflare-workers-ai | 2288 ms | 2288 ms | 108 tok/s n=1 | 100.00% | — | 1 |
| alibaba | 2435 ms | 2435 ms | 94 tok/s n=1 | 100.00% | — | 1 |
| gmi | 2575 ms | 3260 ms | 67 tok/s n=1 | 100.00% | — | 5 |
| streamlake | 2590 ms | 2954 ms | 51 tok/s n=2 | 100.00% | — | 4 |
| baseten | 2634 ms | 4179 ms | 121 tok/s n=1 | 100.00% | — | 2 |
| venice | 2803 ms | 2803 ms | 102 tok/s n=1 | 100.00% | — | 1 |
| arcee | 2826 ms | 3809 ms | 133 tok/s n=1 | 100.00% | — | 6 |
| featherless | 3143 ms | 3143 ms | 62 tok/s n=2 | 100.00% | — | 1 |
| telnyx | 5275 ms | 5275 ms | 192 tok/s n=2 | 100.00% | — | 1 |
| chutes | 13033 ms | 13033 ms | 13 tok/s n=1 | 100.00% | — | 1 |
| atlas-cloud | — | — | 88 tok/s n=2 | — | — | 0 |
| engy | — | — | 241 tok/s n=1 | — | — | 0 |
| inceptron | — | — | — | — | 4 insufficient_quota |
0 |
| siliconflow | — | — | 80 tok/s n=2 | — | — | 0 |
| wandb | — | — | 154 tok/s n=2 | — | — | 0 |
Full provider & model leaderboard.
32 routes.
More routes give the auto router more room to fail over around provider 429 and 5xx responses.
Gateway overhead is measured separately.
Public status separates TLS/health overhead from full model latency so slow LLMs do not inflate the router metric.
Metadata rollups.
Status samples store latency, outcome, provider, model, route, cost, and region metadata only.
View public status or inspect provider routes.