# Fastest AI API from South America (Santiago)
Updated November 08, 2026. Independent, provider-neutral edge latency (TTFB) and uptime by provider, measured from South America (Santiago), updated automatically.

As of November 08, 2026, measured from South America (Santiago), the fastest AI inference API by edge latency (time-to-first-byte) is **openrouter** at 58 ms p50 (n=32).

## Edge latency (time-to-first-byte)
| # | Provider | p50 | p95 | Uptime | Samples |
| --- | --- | --- | --- | --- | --- |
| 1 | openrouter | 58 ms | 83 ms | 100% | 32 |
| 2 | meta-llama | 166 ms | 378 ms | 100% | 32 |
| 3 | typesafe | 174 ms | 179 ms | 100% | 32 |
| 4 | baseten | 178 ms | 198 ms | 100% | 32 |
| 5 | cerebras | 180 ms | 237 ms | 100% | 32 |
| 6 | xai | 190 ms | 233 ms | 100% | 32 |
| 7 | groq | 194 ms | 227 ms | 100% | 32 |
| 8 | anthropic | 195 ms | 235 ms | 100% | 32 |
| 9 | cohere | 196 ms | 226 ms | 100% | 32 |
| 10 | openai | 210 ms | 236 ms | 100% | 32 |
| 11 | inference-net | 213 ms | 292 ms | 100% | 32 |
| 12 | novita | 226 ms | 332 ms | 100% | 32 |
| 13 | perplexity | 230 ms | 481 ms | 100% | 32 |
| 14 | together | 235 ms | 619 ms | 100% | 32 |
| 15 | fireworks | 260 ms | 276 ms | 100% | 32 |
| 16 | reka | 264 ms | 536 ms | 100% | 32 |
| 17 | writer | 268 ms | 329 ms | 100% | 32 |
| 18 | siliconflow | 280 ms | 284 ms | 100% | 32 |
| 19 | mistral | 293 ms | 335 ms | 100% | 32 |
| 20 | friendli | 294 ms | 364 ms | 100% | 32 |
| 21 | kimi | 344 ms | 568 ms | 100% | 32 |
| 22 | ai21 | 376 ms | 498 ms | 100% | 32 |
| 23 | sambanova | 380 ms | 384 ms | 100% | 32 |
| 24 | baichuan | 395 ms | 1385 ms | 100% | 32 |
| 25 | ernie | 414 ms | 423 ms | 100% | 32 |
| 26 | nebius | 432 ms | 471 ms | 100% | 32 |
| 27 | nscale | 441 ms | 444 ms | 100% | 32 |
| 28 | deepseek | 450 ms | 497 ms | 100% | 32 |
| 29 | aleph-alpha | 464 ms | 475 ms | 100% | 32 |
| 30 | minimax | 496 ms | 512 ms | 100% | 32 |
| 31 | glm | 533 ms | 545 ms | 100% | 32 |
| 32 | deepinfra | 537 ms | 614 ms | 100% | 32 |
| 33 | replicate | 560 ms | 769 ms | 100% | 32 |
| 34 | upstage | 593 ms | 686 ms | 100% | 32 |
| 35 | sarvam | 671 ms | 689 ms | 100% | 32 |
| 36 | stepfun | 709 ms | 1729 ms | 100% | 32 |
| 37 | sensenova | 730 ms | 1664 ms | 100% | 32 |
| 38 | qwencloud | 730 ms | 751 ms | 100% | 26 |
| 39 | iflytek | 737 ms | 1674 ms | 100% | 32 |
| 40 | qwen | 742 ms | 773 ms | 100% | 32 |
| 41 | featherless | 746 ms | 891 ms | 100% | 32 |
| 42 | hyperbolic | 754 ms | 782 ms | 100% | 32 |
| 43 | doubao | 762 ms | 1725 ms | 100% | 32 |
| 44 | tencent-tokenhub | 791 ms | 804 ms | 100% | 32 |
| 45 | google | 972 ms | 1071 ms | 100% | 32 |


Source: https://llmlatency.dev/region/sa-west