# Self-hosted AI agents: which VPS region is closest to the LLM APIs

Updated November 07, 2026.

As of November 07, 2026, the median LLM API answered its first byte in 221 ms from Europe; 175 ms from US Central; 208 ms from Tokyo; 337 ms from São Paulo. These are measured from rented VPSes of the kind you would run OpenClaw, Hermes Agent or n8n on — Hetzner in Nuremberg, Vultr in Chicago, Vultr in Tokyo, Vultr in São Paulo — against 43 APIs every five minutes. US Central is closest to the most of the 24 APIs built into these agents (11 answer fastest from there), but the API you use decides.

## Median time to first byte by probe region (24 h)

| Probe region | VPS | Median API | Under 100 ms | Fastest API |
| --- | --- | --- | --- | --- |
| Europe | Hetzner · Nuremberg, Germany | 221 ms | 7 of 43 | Fireworks 16 ms |
| US Central | Vultr · Chicago, US | 175 ms | 14 of 43 | Fireworks 29 ms |
| Tokyo | Vultr · Tokyo, Japan | 208 ms | 6 of 43 | Fireworks 19 ms |
| São Paulo | Vultr · São Paulo, Brazil | 337 ms | 1 of 43 | OpenRouter 62 ms |

## LLM APIs built into OpenClaw, Hermes Agent and n8n, by region

| LLM API | OpenClaw | Hermes Agent | n8n | Europe | US Central | Tokyo | São Paulo |
| --- | --- | --- | --- | --- | --- | --- | --- |
| OpenRouter | yes | yes | yes | 98 ms | 59 ms | 54 ms | 62 ms |
| Fireworks | yes | yes | - | 16 ms | 29 ms | 19 ms | 254 ms |
| Google Gemini | yes | yes | yes | 90 ms | 41 ms | 63 ms | 137 ms |
| Baseten | yes | - | - | 125 ms | 47 ms | 160 ms | 154 ms |
| Cerebras | yes | - | - | 195 ms | 73 ms | 201 ms | 190 ms |
| Anthropic | yes | yes | yes | 197 ms | 95 ms | 227 ms | 194 ms |
| NovitaAI | yes | yes | - | 231 ms | 93 ms | 147 ms | 247 ms |
| xAI | yes | yes | yes | 221 ms | 110 ms | 165 ms | 224 ms |
| Groq | yes | - | yes | 213 ms | 126 ms | 161 ms | 232 ms |
| Together | yes | - | - | 232 ms | 127 ms | 162 ms | 261 ms |
| OpenAI | yes | yes | yes | 215 ms | 139 ms | 208 ms | 244 ms |
| Mistral | yes | - | yes | 117 ms | 196 ms | 322 ms | 264 ms |
| Moonshot (Kimi) | yes | yes | yes | 309 ms | 266 ms | 130 ms | 341 ms |
| Nebius Token Factory | - | yes | - | 33 ms | 181 ms | 466 ms | 377 ms |
| Z.AI (GLM) | yes | yes | - | 184 ms | 252 ms | 94 ms | 541 ms |
| Qianfan (Baidu) | yes | - | - | 286 ms | 272 ms | 138 ms | 412 ms |
| DeepSeek | yes | yes | yes | 290 ms | 317 ms | 183 ms | 432 ms |
| MiniMax | yes | yes | yes | 312 ms | 343 ms | 223 ms | 478 ms |
| Upstage | - | yes | - | 586 ms | 355 ms | 59 ms | 616 ms |
| DeepInfra | yes | - | - | 507 ms | 157 ms | 447 ms | 533 ms |
| Qwen Cloud | yes | yes | yes | 358 ms | 415 ms | 167 ms | 706 ms |
| StepFun | yes | yes | - | 429 ms | 392 ms | 325 ms | 673 ms |
| Featherless AI | yes | - | - | 588 ms | 284 ms | 426 ms | 685 ms |
| Volcano Engine (Doubao) | yes | - | - | 350 ms | 457 ms | 494 ms | 706 ms |

## What is the best VPS region for OpenClaw, Hermes Agent or n8n?

The region where the LLM API you use answers fastest. As of November 07, 2026, of the APIs built into these agents, US Central is closest to the most (11); by API: US Central for Google Gemini, Baseten, Cerebras, Anthropic and 7 more; Tokyo for OpenRouter, Moonshot (Kimi), Z.AI (GLM), Qianfan (Baidu) and 5 more; Europe for Fireworks, Mistral, Nebius Token Factory, Volcano Engine (Doubao).

## Does the VPS location affect AI agent latency?

Yes, on every call. Across the APIs built into these agents, the median gap between an API's fastest and slowest probe region is 225 ms of time to first byte; for Upstage it is 557 ms (59 ms from Tokyo against 616 ms from São Paulo). Over 100 sequential model calls the slower region costs about 22 s of pure network time for a typical API and 56 s for Upstage.

## Which servers do these measurements come from?

Rented VPSes: Europe — Hetzner, Nuremberg, Germany; US Central — Vultr, Chicago, US; Tokyo — Vultr, Tokyo, Japan; São Paulo — Vultr, São Paulo, Brazil. A VPS elsewhere will not match the numbers exactly; the comparison between regions is what carries over.

Per agent: https://llmlatency.dev/self-hosted-agents/openclaw · https://llmlatency.dev/self-hosted-agents/hermes-agent · https://llmlatency.dev/self-hosted-agents/n8n

Source: https://llmlatency.dev/self-hosted-agents · CC BY 4.0