PERFORMANCEGLM-5.2 → Scaleway: TTFT ↑ 20%27mPERFORMANCEDeepSeek V4 Flash → Scaleway: TTFT ↑ 35%27mPERFORMANCEMistral Medium 3.5 → Scaleway: TTFT ↓ 21%27mPERFORMANCEDeepSeek V4 Flash → Scaleway: throughput ↓ 20%27mPERFORMANCEGPT-OSS 120B → Scaleway: throughput ↑ 31%27mPERFORMANCEGPT-OSS 120B → Scaleway: TTFT ↓ 24%27mPERFORMANCEGPT-OSS 120B → Together AI: throughput ↑ 20%27mPERFORMANCEKimi K3 via OpenRouter: TTFT ↓ 49%27mPERFORMANCEGLM-5.2 via OpenRouter: reliability recovered 96.8% → 100.0%27mPERFORMANCEKimi K3 via Cortecs: throughput ↑ 184%27mPERFORMANCEKimi K3 via Cortecs: TTFT ↓ 42%27mPERFORMANCEGLM-5.2 via OpenRouter: throughput ↓ 49%27mPERFORMANCEGLM-5.2 via OpenRouter: TTFT ↑ 121%27mPERFORMANCEMiniMax M3 via OpenRouter: reliability recovered 96.8% → 100.0%27mPERFORMANCEDeepSeek V4 Pro via Cortecs: throughput ↑ 27%27mPERFORMANCEGPT-OSS 120B via OpenRouter: throughput ↓ 22%27mPERFORMANCEGLM-4.7 via OpenRouter: TTFT ↓ 78%27mPERFORMANCEGPT-OSS 20B → Groq: TTFT ↑ 24%27mPERFORMANCEKimi K3 → Together AI: TTFT ↓ 34%27mPERFORMANCEDeepSeek V4 Flash via OpenRouter: throughput ↓ 17%27mPERFORMANCELlama 3.3 70B via Cortecs: TTFT ↓ 51%27mPERFORMANCEQwen3 235B via OpenRouter: throughput ↑ 37%27mPERFORMANCEQwen3 235B via OpenRouter: TTFT ↓ 29%27mPERFORMANCEDeepSeek V4 Flash → Together AI: throughput ↑ 19%27mPERFORMANCEGLM-5.3 via OpenRouter: throughput ↓ 17%27mPERFORMANCEGPT-OSS 20B via Cortecs: throughput ↑ 28%27mPERFORMANCELlama 3.3 70B → Together AI: TTFT ↓ 33%27mPRICE~moonshotai/kimi-latest cached_input_per_mtok decreased from 0.80 to 0.2927mPRICE~moonshotai/kimi-latest output_per_mtok decreased from 13.00 to 11.3627mPRICE~moonshotai/kimi-latest input_per_mtok decreased from 0.99 to 0.7527mPRICE~deepseek/deepseek-v4-flash-latest cached_input_per_mtok decreased from 0.0077 to 0.001127mPRICE~deepseek/deepseek-v4-flash-latest output_per_mtok decreased from 1.28 to 1.0427mPRICE~deepseek/deepseek-v4-flash-latest input_per_mtok decreased from 0.0077 to 0.003827mPRICEz-ai/glm-5.1 cached_input_per_mtok increased from 0.18 to 0.2627mPRICEz-ai/glm-5.1 output_per_mtok increased from 3.03 to 4.4027mPRICEz-ai/glm-5.1 input_per_mtok increased from 0.96 to 1.4027mPRICEtencent/hy4-preview cached_input_per_mtok increased from 0.038 to 0.04227mPRICEtencent/hy4-preview input_per_mtok increased from 0.75 to 0.8327mPRICEtencent/hy3 cached_input_per_mtok increased from 0.021 to 0.03327mPRICEtencent/hy3 output_per_mtok increased from 0.33 to 0.5327m

AI inference providers

The global landscape of inference infrastructure — 92 providers in the registry, live from continuous discovery.

Measured independently

best measured path per provider · canonical vantage

Registry census

catalogued daily by discovery — benchmarks begin when measurement credentials land
ProviderHQTypeEndpointsState
Alibaba Cloud Model StudioCNhyperscaler2discovered
AWS BedrockUShyperscaler2discovered
AnthropicUSfirst party model lab1benchmarked
DeepInfraUSinference host1discovered
DeepSeekCNfirst party model lab1discovered
Fireworks AIUSinference host1benchmarked
HelmcodeESinference host1discovered
Infomaniak AI ServicesCHsovereign cloud1discovered
IONOS AI Model HubDEsovereign cloud1discovered
Moonshot AICNfirst party model lab1discovered
Nebius Token FactoryNLinference host1benchmarked
NextBitESinference host1discovered
NscaleGBinference host1discovered
OVHcloudFRsovereign cloud1discovered
RegoloITinference host1discovered
SiliconFlowCNinference host1discovered
AI21 LabsILfirst party model lab—discovered
AIMLAPIEErouter—discovered
Aion LabsILfirst party model lab—discovered
AkashMLUSinference host—discovered
AKI.IODEinference host—discovered
Aleph AlphaDEfirst party model lab—discovered
Arcee AIUSfirst party model lab—discovered
AtlasCloudUSinference host—discovered
Baichuan AICNfirst party model lab—discovered
Baidu QianfanCNhyperscaler—discovered
BasetenUSmanaged endpoint platform—discovered
ByteDance Volcano EngineCNhyperscaler—discovered
ChutesUSinference host—discovered
Cloudflare AI GatewayUSai gateway—discovered
Cloudflare Workers AIUShyperscaler—discovered
CohereCAfirst party model lab—discovered
CoreWeaveUSinference host—discovered
CrusoeUSinference host—discovered
DecartILfirst party model lab—discovered
DigitalOcean AIUShyperscaler—discovered
EUrouterDErouter—discovered
FriendliAIKRinference host—discovered
GMI CloudUSinference host—discovered
Google AI StudioUSfirst party model lab—discovered
Google Vertex AIUShyperscaler—discovered
Huawei Cloud ModelArtsCNhyperscaler—discovered
Hugging Face Inference ProvidersUSai gateway—discovered
HyperbolicUSinference host—discovered
InceptronSEinference host—discovered
Infercom.aiLUinference host—discovered
InworldUSinference host—discovered
io.netUSinference host—discovered
IonstreamUSinference host—discovered
Lyceum TechnologyDEinference host—discovered
MeliousDEinference host—discovered
MetaUSfirst party model lab—discovered
Microsoft FoundryUShyperscaler—discovered
MiniMaxCNfirst party model lab—discovered
ModalUSmanaged endpoint platform—discovered
MorphUSinference host—discovered
NebulNLsovereign cloud—discovered
Nex AGICNinference host—discovered
NovitaAIUSinference host—discovered
NVIDIA NIMUShyperscaler—discovered
OnshoreAIGBinference host—discovered
OpenInferenceUSinference host—discovered
ParasailUSinference host—discovered
PerplexityUSfirst party model lab—discovered
PhalaUSinference host—discovered
PoolsideUSfirst party model lab—discovered
PortkeyUSai gateway—discovered
Reka AIUSfirst party model lab—discovered
ReplicateUSinference host—discovered
RunPodUSmanaged endpoint platform—discovered
Sakana AIJPfirst party model lab—discovered
SambaNovaUSinference host—discovered
SourcefulSEinference host—discovered
StepFunCNfirst party model lab—discovered
StreamLakeCNinference host—discovered
Tencent Cloud TokenHubCNhyperscaler—discovered
TensorixIEinference host—discovered
UpstageKRfirst party model lab—discovered
VeniceUSinference host—discovered
Vercel AI GatewayUSai gateway—discovered
WaferUSinference host—discovered
xAIUSfirst party model lab—discovered
Xiaomi MiMoCNfirst party model lab—discovered
Zhipu AICNfirst party model lab—discovered