MoonshotAI: Kimi K3 Performance
Try Kimi K3 on us. Starter credit covers your first requests, then one key reaches all 28 provider routes with automatic failover.
moonshotai/kimi-k3
Measured performance
Continuously sampled p50/p95 time-to-first-token (TTFT), effective throughput, and success rate for MoonshotAI: Kimi K3. 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 |
|---|---|---|---|---|---|---|
| sail-research | 1674 ms | 1837 ms | 132 tok/s n=1 | 66.67% | — | 12 |
| relace | 1882 ms | 8072 ms | 57 tok/s n=1 | 100.00% | — | 15 |
| kimi | 8848 ms | 13618 ms | — | 100.00% | — | 10 |
| together | 838 ms | 4545 ms | 34 tok/s n=1 | 100.00% | — | 2 |
| tinfoil | 996 ms | 1117 ms | 89 tok/s n=1 | 100.00% | — | 2 |
| telnyx | 1052 ms | 2483 ms | 139 tok/s n=2 | 100.00% | — | 2 |
| engy | 1272 ms | 2159 ms | 59 tok/s n=2 | 100.00% | — | 6 |
| makora | 1999 ms | 2153 ms | 29 tok/s n=1 | 100.00% | — | 5 |
| akashml | 2206 ms | 11914 ms | 139 tok/s n=1 | 100.00% | — | 8 |
| tencent | 2538 ms | 4195 ms | 33 tok/s n=2 | 100.00% | — | 3 |
| cloudflare-workers-ai | 2884 ms | 3889 ms | 45 tok/s n=2 | 100.00% | — | 2 |
| gmi | 2991 ms | 20736 ms | 104 tok/s n=2 | 100.00% | — | 3 |
| siliconflow | 3021 ms | 6095 ms | 31 tok/s n=1 | 100.00% | — | 2 |
| novita | 3101 ms | 3101 ms | 23 tok/s n=2 | 100.00% | — | 1 |
| phala | 3147 ms | 18611 ms | 22 tok/s n=2 | 100.00% | 1 provider_error |
8 |
| arcee | 3763 ms | 4584 ms | 86 tok/s n=2 | 100.00% | — | 6 |
| fireworks | 4252 ms | 4252 ms | 67 tok/s n=1 | 100.00% | — | 1 |
| featherless | 5544 ms | 13223 ms | 42 tok/s n=2 | 100.00% | — | 3 |
| deepinfra | — | — | — | 100.00% | — | 330 |
| atlas-cloud | — | — | 35 tok/s n=2 | — | — | 0 |
| baseten | — | — | 71 tok/s n=1 | — | — | 0 |
| boundless | — | — | 101 tok/s n=1 | — | — | 0 |
| chutes | — | — | 20 tok/s n=1 | 0.00% | — | 2 |
| io-net | — | — | 94 tok/s n=2 | — | — | 0 |
| lyceum | — | — | 145 tok/s n=1 | — | — | 0 |
| nebius | — | — | 78 tok/s n=1 | — | — | 0 |
| venice | — | — | 70 tok/s n=1 | — | — | 0 |
Full provider & model leaderboard.
28 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.