# Fastest AI API from US (Central)
Updated July 23, 2026. Independent, provider-neutral edge latency (TTFB) and uptime by provider, measured from US (Central), updated automatically.

As of July 23, 2026, measured from US (Central), the fastest AI inference API by edge latency (time-to-first-byte) is **fireworks** at 23 ms p50 (n=58).

## Edge latency (time-to-first-byte)
| # | Provider | p50 | p95 | Uptime | Samples |
| --- | --- | --- | --- | --- | --- |
| 1 | fireworks | 23 ms | 65 ms | 100% | 58 |
| 2 | google | 34 ms | 93 ms | 100% | 58 |
| 3 | meta-llama | 54 ms | 115 ms | 100% | 35 |
| 4 | openrouter | 54 ms | 102 ms | 100% | 58 |
| 5 | baseten | 56 ms | 68 ms | 100% | 35 |
| 6 | novita | 59 ms | 137 ms | 100% | 35 |
| 7 | cohere | 61 ms | 112 ms | 100% | 58 |
| 8 | cerebras | 66 ms | 144 ms | 100% | 58 |
| 9 | replicate | 75 ms | 116 ms | 100% | 35 |
| 10 | friendli | 80 ms | 339 ms | 100% | 35 |
| 11 | siliconflow | 86 ms | 106 ms | 100% | 35 |
| 12 | targon | 90 ms | 98 ms | 100% | 35 |
| 13 | perplexity | 107 ms | 145 ms | 100% | 35 |
| 14 | anthropic | 108 ms | 264 ms | 100% | 58 |
| 15 | openai | 114 ms | 194 ms | 100% | 58 |
| 16 | groq | 119 ms | 184 ms | 100% | 58 |
| 17 | together | 126 ms | 170 ms | 100% | 58 |
| 18 | xai | 126 ms | 838 ms | 100% | 58 |
| 19 | writer | 138 ms | 166 ms | 100% | 35 |
| 20 | ai21 | 157 ms | 236 ms | 100% | 35 |
| 21 | inference-net | 157 ms | 198 ms | 100% | 35 |
| 22 | sambanova | 160 ms | 220 ms | 100% | 35 |
| 23 | hyperbolic | 170 ms | 272 ms | 100% | 35 |
| 24 | mistral | 170 ms | 252 ms | 100% | 58 |
| 25 | nscale | 196 ms | 209 ms | 100% | 35 |
| 26 | nebius | 223 ms | 311 ms | 100% | 35 |
| 27 | deepinfra | 227 ms | 403 ms | 100% | 35 |
| 28 | baichuan | 235 ms | 879 ms | 100% | 35 |
| 29 | kimi | 266 ms | 308 ms | 100% | 40 |
| 30 | ernie | 271 ms | 277 ms | 100% | 35 |
| 31 | aleph-alpha | 301 ms | 365 ms | 100% | 35 |
| 32 | deepseek | 305 ms | 355 ms | 100% | 58 |
| 33 | upstage | 316 ms | 334 ms | 100% | 35 |
| 34 | minimax | 324 ms | 351 ms | 100% | 40 |
| 35 | featherless | 365 ms | 1356 ms | 100% | 35 |
| 36 | yi-01ai | 373 ms | 1060 ms | 100% | 35 |
| 37 | glm | 389 ms | 418 ms | 100% | 40 |
| 38 | reka | 402 ms | 488 ms | 100% | 35 |
| 39 | sarvam | 436 ms | 464 ms | 100% | 35 |
| 40 | stepfun | 436 ms | 1130 ms | 100% | 35 |
| 41 | qwen | 437 ms | 491 ms | 100% | 40 |
| 42 | sensenova | 443 ms | 1089 ms | 100% | 35 |
| 43 | iflytek | 443 ms | 3195 ms | 97% | 35 |
| 44 | doubao | 519 ms | 1533 ms | 100% | 35 |
| 45 | hunyuan | 1379 ms | 1387 ms | 100% | 35 |


Source: https://llmlatency.dev/region/us-central