Data as of · 78 models, 80 questions · reviewed by LLM Verifications team · methodology

Hosts · catalogue checked 27 Oct 2026

LLM API providers compared

70 API providers (hosts) serve the 78 models we test through OpenRouter. Compare who serves what, how their list prices compare with the cheapest host for the same model, and their recent uptime.

All hosts

Price premium: median markup over the cheapest host, across models served by two or more hosts

HostModelsCheapest forMedian premiumUptime (30 min)Largest contextTop model served
DeepInfra267 of 23+24%99.4%1MDeepSeek V4.1 Flash 1017
Novita192 of 19+54%100%1MDeepSeek V4.1 Flash 1017
Parasail160 of 15+87%97.6%1MDeepSeek V4.1 Flash 1017
Amazon Bedrock154 of 13+120%87.0%1.1MClaude Opus 5.5
CoreWeave152 of 15+73%99.6%1MDeepSeek V4 Pro 0813
Google157 of 14+41%75.7%1.3MClaude Opus 5.5
Venice152 of 15+69%97.3%1MDeepSeek V4 Pro 0813
Azure135 of 13+100%90.9%1.1MClaude Opus 5.5
SiliconFlow132 of 13+167%92.6%1MDeepSeek V4.1 Flash 1017
GMICloud122 of 12+64%100%1.1MDeepSeek V4.1 Flash 1017
OpenAI119 of 9Cheapest82.7%1.1MGPT-5.3-Codex
Together112 of 10+208%100%1MMuse Spark Open
Alibaba103 of 8+99%100%1MQwen3.8 Max (0902)
AtlasCloud100 of 9+142%100%1MDeepSeek V4.1 Flash 1017
Nebius101 of 9+337%100%1MMuse Spark Open
Phala90 of 9+297%100%1MDeepSeek V4.1 Flash 1017
BaseTen81 of 8+298%100%1MMuse Spark Open
Cloudflare80 of 7+405%100%1MDeepSeek V4.1 Flash 1017
Darkbloom86 of 8Cheapest100%1MGLM 5.3 Flash
DekaLLM80 of 8+55%99.6%1MDeepSeek V4 Pro 0813
DigitalOcean81 of 8+199%100%1MDeepSeek V4.1 Flash 1017
Mistral80 of 2+420%100%1MGLM 5.3 Flash
StreamLake80 of 8+48%92.9%1MDeepSeek V4.1 Flash 1017
AkashML70 of 7+41%100%1MGLM 5.3
Crusoe60 of 6+212%100%1MDeepSeek V4.1 Flash 1017
Makora61 of 6+115%100%1MMuse Spark Open
InferenceNet51 of 5+74%100%1MDeepSeek V4.1 Flash 1017
Io Net51 of 5+68%98.6%1MMiMo-V2.6-Flash
Modal51 of 5+286%100%1MMuse Spark Open
ModelRun50 of 5+356%75.0%1MGLM 5.3
Reka50 of 4+149%100%1MDeepSeek V4.1 Flash 1017
Relace51 of 5+57%100%1MDeepSeek V4.1 Flash 1017
Sail Research51 of 5+142%98.9%1MDeepSeek V4.1 Flash 1017
Wafer51 of 5+93%100%1MDeepSeek V4.1 Flash 1017
Anthropic44 of 4Cheapest100%1MClaude Opus 5.5
Fireworks41 of 4+249%100%1MMuse Spark Open
Google AI Studio44 of 4Cheapest100%1MGemini 4 Argon
PrimeIntellect40 of 4+319%100%1MDeepSeek V4.1 Flash 1017
SambaNova40 of 4+260%100%1MMiniMax M3.1 Flash
Baidu31 of 3+335%100%1MDeepSeek V4.1 Flash 1017
Claude Platform on AWS33 of 3Cheapest100%1MClaude Opus 5.5
Cohere30 of 2+213%100%1MGLM 5.3 Flash
Friendli30 of 3+381%100%1MGLM 5.3 Flash
Groq30 of 3+313%100%131kgpt-oss-120b
Inceptron31 of 3+341%100%1MGLM 5.3 Flash
Ionstream30 of 3+204%100%1MDeepSeek V4.1 Flash 1017
Morph30 of 3+331%100%1MDeepSeek V4.1 Flash 1017
Near AI31 of 3+116%100%1MGLM 5.3 Flash
Cerebras20 of 2+603%100%131kQwen3.8 27B
Chutes20 of 2+58%100%262kQwen3.8 27B
Decart20 of 2+162%100%1MDeepSeek V4.1 Flash 1017
DeepSeek20 of 2+252%100%1MDeepSeek V4.1 Flash 1017
Mancer 220 of 2+207%100%262kQwen3.8 27B
Mara20 of 2+285%96.7%1Mgpt-oss-120b
Minimax20 of 2+20%100%1MMiniMax M3.1 Flash
Moonshot AI20 of 1+43%100%1MKimi K3.1 Pro
NextBit20 of 2+215%100%1MGLM 5.3 Flash
OpenInference20 of 2+99%100%1MDeepSeek V4.1 Flash 1017
Tenstorrent21 of 2+84%100%1MGLM 5.3
Upstage2——100%524kSolar Pro 4
Xiaomi20 of 2+19%100%1MMiMo-V2.6-Pro
Z.AI20 of 2+420%100%1MGLM 5.3 Flash
Arcee AI1——100%262kTrinity Large Thinking
Featherless11 of 1Cheapest—262kStep 5 Preview
Inception1——100%260kMercury 2.5
Liquid1———128k—
Meta1——100%1MMuse Wonder
StepFun11 of 1Cheapest—1MStep 5 Preview
Tencent11 of 1Cheapest100%262kHy3
xAI1——100%500kGrok 4.7

Frequently asked questions

Which API host serves the most AI models?
DeepInfra serves 26 of the 78 models we test, according to OpenRouter's endpoint list.
Which API host is cheapest?
Among hosts that compete on at least three models, OpenAI is cheapest most often: the lowest list price for 9 of the 9 models it shares with other hosts.
Do hosts charge different prices for the same model?
For 35 of the 47 models served by two or more hosts, list prices differ between hosts. The widest gap is DeepSeek V4 Pro 0813: 11.8× from the cheapest host (Baidu) to the most expensive (Venice), by blended price.
Is speed measured per host?
Not yet. Our speed runs went through whichever host OpenRouter routed each request to, so host pages compare price, uptime and limits, not speed.