# Best LLM API for OpenClaw, measured

Updated November 07, 2026. Built-in providers checked against OpenClaw's documentation (https://docs.openclaw.ai/concepts/model-providers/official-provider-plugins) on 2026-11-07.

As of November 07, 2026, of the 22 LLM APIs built into OpenClaw that we measure, OpenRouter answers fastest: 68 ms median time to first byte, averaged over 4 regions. By region, Fireworks is fastest from Europe (16 ms), US Central (29 ms) and Tokyo (19 ms), OpenRouter from São Paulo (62 ms). 6 of 22 hosts answered every probe over the 90 days since 2026-08-09; the lowest was OpenAI at 99.1%.

Median time to first byte (24 h) per region, host uptime over 90 days since 2026-08-09, outages attributed to the provider in the last 30 days (host, and real requests to one model where we hold a key). Measured at the exact host OpenClaw calls, directly.

| Provider | Id | Host | Europe | US Central | Tokyo | São Paulo | Host uptime 90 d | Outages 30 d |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| OpenRouter | `openrouter` | openrouter.ai | 98 ms | 59 ms | 54 ms | 62 ms | 100% | 0 |
| Fireworks | `fireworks` | api.fireworks.ai | 16 ms | 29 ms | 19 ms | 254 ms | 100% | 0 |
| Google Gemini | `google` | generativelanguage.googleapis.com | 90 ms | 41 ms | 63 ms | 137 ms | 100% | 119 requests to gemini-flash-lite-latest |
| Baseten | `baseten` | inference.baseten.co | 125 ms | 47 ms | 160 ms | 154 ms | 99.9% | 0 |
| Cerebras | `cerebras` | api.cerebras.ai | 195 ms | 73 ms | 201 ms | 190 ms | 99.9% | 1 host; 1 request to gpt-oss-120b |
| Anthropic | `anthropic` | api.anthropic.com | 197 ms | 95 ms | 227 ms | 194 ms | 100% | 0 |
| NovitaAI | `novita` | api.novita.ai | 231 ms | 93 ms | 147 ms | 247 ms | 99.9% | 0 |
| xAI | `xai` | api.x.ai | 221 ms | 110 ms | 165 ms | 224 ms | 99.9% | 0 |
| Groq | `groq` | api.groq.com | 213 ms | 126 ms | 161 ms | 232 ms | 100% | 0 |
| Together | `together` | api.together.xyz | 232 ms | 127 ms | 162 ms | 261 ms | 100% | 0 |
| OpenAI | `openai` | api.openai.com | 215 ms | 139 ms | 208 ms | 244 ms | 99.1% | 0 |
| Mistral | `mistral` | api.mistral.ai | 117 ms | 196 ms | 322 ms | 264 ms | 99.9% | 0 |
| Moonshot (Kimi) | `moonshot` | api.moonshot.ai (international endpoint) | 309 ms | 266 ms | 130 ms | 341 ms | 99.9% | 0 |
| Z.AI (GLM) | `zai` | api.z.ai | 184 ms | 252 ms | 94 ms | 541 ms | 99.9% | 1 host |
| Qianfan (Baidu) | `qianfan` | qianfan.baidubce.com (China endpoint) | 286 ms | 272 ms | 138 ms | 412 ms | 99.9% | 0 |
| DeepSeek | `deepseek` | api.deepseek.com | 290 ms | 317 ms | 183 ms | 432 ms | 99.9% | 0 |
| MiniMax | `minimax` | api.minimaxi.com (China endpoint; the global api.minimax.io is not measured) | 312 ms | 343 ms | 223 ms | 478 ms | 99.9% | 0 |
| DeepInfra | `deepinfra` | api.deepinfra.com | 507 ms | 157 ms | 447 ms | 533 ms | 99.8% | 0 |
| Qwen Cloud | `qwen` | dashscope-intl.aliyuncs.com (international endpoint) | 358 ms | 415 ms | 167 ms | 706 ms | 99.9% | 0 |
| StepFun | `stepfun` | api.stepfun.com (China endpoint; the global api.stepfun.ai is not measured) | 429 ms | 392 ms | 325 ms | 673 ms | 99.7% | 2 host |
| Featherless AI | `featherless` | api.featherless.ai | 588 ms | 284 ms | 426 ms | 685 ms | 99.8% | 0 |
| Volcano Engine (Doubao) | `volcengine` | ark.cn-beijing.volces.com (China endpoint) | 350 ms | 457 ms | 494 ms | 706 ms | 99.3% | 20 host |

Not shown — the agent calls a host we do not measure:

- Cohere: OpenClaw calls api.cohere.ai, our probe measures api.cohere.com (https://docs.openclaw.ai/providers/cohere)
- Meta: OpenClaw calls api.meta.ai, our probe measures api.llama.com (https://docs.openclaw.ai/providers/meta)
- Tencent TokenHub: OpenClaw calls tokenhub.tencentmaas.com, our probe measures api.hunyuan.cloud.tencent.com, the Tencent Hunyuan API, shut down on September 30, 2026 (https://docs.openclaw.ai/providers/tencent)

Any other OpenAI-compatible API: a custom provider under `models.providers` with its own base URL (https://docs.openclaw.ai/concepts/model-providers/custom-providers).

## What is the best LLM API for OpenClaw?

It depends on what the agent waits for and where it runs. On measured speed, as of November 07, 2026 the fastest of the 22 built-in APIs we probe is OpenRouter at 68 ms median time to first byte across 4 regions; by region, Fireworks is fastest from Europe (16 ms), US Central (29 ms) and Tokyo (19 ms), OpenRouter from São Paulo (62 ms). On reliability, 6 of 22 hosts answered every probe over the last 90 days. Model quality and price are separate questions this measurement does not answer.

## Which LLM providers does OpenClaw support out of the box?

OpenClaw's documentation lists built-in providers; 22 of them are measured here at the exact host OpenClaw calls: OpenRouter, Fireworks, Google Gemini, Baseten, Cerebras, Anthropic, NovitaAI, xAI, Groq, Together, OpenAI, Mistral, Moonshot (Kimi), Z.AI (GLM), Qianfan (Baidu), DeepSeek, MiniMax, DeepInfra, Qwen Cloud, StepFun, Featherless AI, Volcano Engine (Doubao). Any other OpenAI-compatible API can be added as a custom provider under models.providers with its own base URL.

## Is there a free LLM API for OpenClaw?

Yes, with limits. Google Gemini: free input and output tokens on a limited set of Gemini models (rate-limited; content sent on the free tier is used to improve Google's products; apps for users in the EEA, Switzerland or the UK must use paid services (terms: https://aistudio.google.com/docs/terms)). OpenRouter: model variants whose ID ends in :free (20 requests a minute; 50 a day, or 1,000 a day once you have bought at least 10 credits). Groq: a Free plan; a payment method is needed only to move up to the Developer tier (per-model rate limits, published on Groq's rate-limits page). Z.AI (GLM): GLM-4.7-Flash, GLM-4.5-Flash and GLM-4.6V-Flash: input and output both free (rate limits not published in the docs; the other GLM models are paid). NovitaAI: one model priced at $0 for input and output: Apodex 1.1 Mini (rate limits and how long it stays free are not stated). All are built into OpenClaw; terms checked 2026-11-07 on each provider's own page, full list at https://llmlatency.dev/free-llm-api.

## Does the VPS region matter for OpenClaw?

For some APIs a lot. Qwen Cloud answers in 167 ms from Tokyo and 706 ms from São Paulo, and every model call waits at least that long before the model starts. Put the agent where the API you use is fastest.

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