All promotions
GuidesMCP

Providers

55 providers · 864 offers · prices read Oct 23, 09:51 UTC

Columns
ProviderModels ↓Cheapest forWith cacheUptime 24hData stays inSign-upBased inMedian premium
Novita AI871656%93.0%not statedNo cardUnited States+30%
DeepInfra823957%99.9%Data may be transferred, stored and processed in the United Statesnot checkedUnited States0%
Alibabamaker53947%100%The region you pick sets data storage; deployment scope sets inference locationCard neededChina+74%
Venice40675%99.4%not statedCard neededUnited States+45%
SiliconFlow39659%99.6%not statedCard neededChina+62%
Parasail35391%100%not statedCard neededUnited States+68%
AtlasCloud291100%99.6%Services hosted in the United StatesCard neededUnited States+43%
Googlemaker28650%66.6%not statedNo cardUnited States+54%
StreamLake26677%100%Data stored on servers in Singaporenot checkedChina+10%
CoreWeave252100%100%not statedCard neededUnited States+66%
GMICloud24596%100%not statedCard neededUnited States+26%
Together AI23370%99.9%not statednot checkedUnited States+68%
Phala22086%99.6%not statedCard neededGlobal+59%
Cloudflare21062%100%not statedNo cardGlobal+114%
DigitalOcean210100%100%not statedCard neededUnited States+106%
Mistral AImaker19179%100%Stored data hosted in the EU by default; global endpoint has no fixed inference locationNo cardFrance+146%
Amazon Bedrock17453%100%Geographic cross-Region profiles (US, EU, APAC) stay within that geographyCard neededUnited States+10%
BaseTen161100%99.9%not statedCard neededUnited States+96%
Nebius16138%98.3%Public endpoints run globally; retained data stored in FinlandCard neededFinland+83%
Reka16288%99.9%not statedCard neededUnited States+75%
Crusoe142100%99.9%not statedCard neededUnited States+106%
Z.AImaker143100%99.8%Personal data generally processed in SingaporeCard neededChina+33%
DekaLLM13077%99.9%Hosted in Indonesia; prompts and outputs stay in Indonesianot checkedIndonesia+41%
Baidu120100%100%not statedCard neededChina+111%
OpenAImaker123100%100%Optional regional processing per project (e.g. US, Europe, Japan)Card neededUnited States0%
Io Net113100%98.8%not statedCard neededUnited States+42%
Relace114100%100%not statedCard neededUnited States+44%
Anthropicmaker104100%100%Global by default; US-only inference available as an option$5 creditsUnited States0%
Fireworks AI101100%98.2%not statednot checkedUnited States+146%
Coheremaker9056%100%not statedNo cardUnited States+160%
Minimaxmaker9367%100%not statedCard neededChina+23%
AkashML8288%100%not statedCard neededGlobal+28%
InferenceNet81100%100%Optional US-only inference endpointCard neededUnited States+103%
Modal82100%100%Shared endpoint requests route through us-westCard neededUnited States+114%
NextBit8175%100%EU endpoint runs inference only in Spain; global endpoint may run outside the EEAnot checkedSpain+79%
Azure OpenAI73100%100%Standard: in your geography; Data Zone: within US or EU; Global: any regionCard neededUnited States+130%
Friendli7086%100%not statedCard neededUnited States+121%
Groq7086%99.9%Retained customer data stored in the United StatesNo cardUnited States+220%
Morph70100%99.6%not statedCard neededUnited States+98%
SambaNova7029%98.6%not statedCard neededUnited States+221%
Chutes64100%99.4%not statedCard neededUnited States0%
Inceptron61100%99.8%EU data residency for model APIs (as stated on its homepage)not checkedFinland+51%
Xiaomimaker50100%100%not statedCard neededSingapore+43%
DeepSeekmaker40100%—Personal data processed and stored in the People's Republic of ChinaNo cardAsia-Pacific+48%
Moonshot AImaker30100%100%Collected information stored on servers in SingaporeCard neededChina+59%
xAImaker32100%—US endpoint keeps inference in the US; global endpoint not guaranteedCard neededUnited States0%
Cerebras2050%99.7%not statedCard neededUnited States+302%
Metamaker2—100%97.8%not statednot checkedUnited States—
Pareto Inference10100%—not statednot checkedUnited States+38%
StepFunmaker11100%100%Servers located in the United Statesnot checkedChina0%
Liquid AIno price yet0———not statednot checkedUnited States—
NVIDIAmakerno price yet0———not statedNo cardUnited States—
OpenRouterno price yet0———EU or US in-region processing on Business and Enterprise plansNo cardUnited States—
Poolsideno price yet0———not statednot checkedUnited States—
Recursiveno price yet0———not statednot checkedUnited States—

Models: priced models it serves · cheapest for: models where it is the cheapest host, among those another host also serves · with cache: share of its offers with a cached-input price · uptime: median over its models, OpenRouter, last 24 h · data and sign-up terms: the provider's own pages, linked on each provider page.

Same weights, different bills

For each model a host shares with other hosts, we compare its price (input + output) with the cheapest host of the same weights. Of the 40 hosts that share 5 models or more, half charge at least +67% over the cheapest host. DeepInfra stays within +0% on 76 shared models; SambaNova charges +221%, ×3.2 the cheapest host. DeepInfra is the cheapest host on 39 of the 76 models it shares, more than any other host; 10 of these 40 hosts are never the cheapest.