AI API Relay Comparison 2026
These reviews are independently written and edited — we currently have no commercial partnership with any provider listed below, and the order reflects our own rating, not a paid placement. Some figures come from providers' own marketing materials; always check the official site for current pricing. Each card shows its own last-verified date.
📌 English rollout in progress: we've translated in-depth reviews for 153 of the 153+ relay providers tracked on the Chinese side of this site, and are adding more over time. Looking for a provider that isn't here yet? Browse the full Chinese-language comparison board →
Provider comparison
Sorted by our editorial rating, highest first.
Kling AI
Kuaishou's official Kling open platform, an industry-benchmark video generation model, full text-to-video/image-to-video coverage
DeepSeek
DeepSeek's official API, with V4-series pricing that upended the market and strong reasoning ability
Zhipu AI GLM
Official Zhipu API — GLM-4.7-Flash permanently free, 200K context + 59.2% SWE-bench, mainland direct connect
Krea AI
Proprietary Krea 2 models plus aggregated Nano Banana/Seedance/Kling/Veo access, $83M raised, led by a16z and Bain
Groq Cloud
Powered by purpose-built LPU chips, among the fastest inference speeds in the world, a generous free tier for open-source models
Helicone
0% markup pass-through, built-in LLM observability, open-source and self-hostable, full logging/cost/session tracking
nexos.ai
Founded by the Nord Security team, €30M Series A, an enterprise security/compliance AI gateway
PixVerse
AIsphere's short-video generation model, a unicorn valued at over $2B, backed by Alibaba and CDH
Runware
Lowest-priced image/video generation API on the market, $50M Series A, millisecond-scale inference infrastructure
Alibaba Cloud Bailian
Alibaba Cloud's official AI platform, enterprise-grade SLA, Alibaba's proprietary Qwen model series
Cloudflare AI Gateway
Free AI traffic control layer — caching, rate limiting, monitoring, edge acceleration
Fal AI
Ultra-fast Flux-series image generation, LoRA fine-tuning, the top choice for creative AI workflows
LingyaAI
Mainland dedicated line, 600+ models unified, corporate VAT invoicing available — built for long-running production and enterprise compliance
Moonshot AI (Kimi)
Kimi's official API — 256K ultra-long context, standout document processing and coding ability
NoneLinear
480+ models, 99.99% SLA, the top choice for production-grade enterprise use
OpenRouter
A cross-vendor multi-model aggregator — 400+ models across 70+ providers, one API key to call OpenAI/Anthropic/Google/Meta/Mistral and more
Portkey
Now part of Palo Alto Networks — the core AI gateway for Prisma AIRS; unified access to 1600+ models, full LLMOps suite
SiliconFlow
Domestic open-source model cloud hosting — DeepSeek, Qwen, GLM and 100+ models, an in-house inference engine plus domestic chip support
Together AI
Large-scale open-source model compute aggregation — 100+ models behind one API, suited to both research and enterprise use
TokenRiver
Enterprise-grade AI large-model intelligent gateway, 650+ models, ¥1=$1 exchange-rate advantage, multi-channel automatic failover, redirect destination for the former ShiyunApi
Vercel AI Gateway
0% markup, $5 free credit, zero data retention (ZDR), the top choice for frontend developers
Fireworks AI
Enterprise-grade inference for open-source models, deep Function Calling optimization, ultra-low-latency SLA
Jina AI API
One key, one token pool for Embeddings + Reranker + Reader — 10M free tokens, no credit card required
Lambda Labs
GPU cloud compute + open-model inference; rebranded as Lambda, 60-70% cheaper than mainstream clouds, a top pick for AI research
Laozhang API
A well-known, long-established Chinese Claude relay, priced at parity with official rates, with a strong reputation for stability — a solid first choice if you don't want to deal with headaches
n1n.ai
Enterprise unified LLM API gateway — one key for 400+ models, ¥20 free credit for new users
Requesty
EU-friendly, 600+ models, GDPR-compliant, 20ms failover
Weelinking
99.99% SLA, 200+ models, multi-layer redundancy — a stable first choice for enterprise production environments
4SAPI / Starlink 4SAPI
Enterprise-grade "self-healing routing system" + enterprise account pool, focused on low latency and high concurrency — frequently tops reviews for AI coding assistants and multimodal use cases
AnPin AI
1Gbps dedicated line + multi-node routing, 99.99% stability promise, admin responds in real time on X
API Yi
Long-established AI API aggregator, 300K monthly visits, broad frontier-model coverage
Zhipu AI (BigModel)
Zhipu's official GLM flagship: GLM-5.2 (1M context) in production, GLM-5.3 coming soon, multimodal AI, top-tier Chinese-language capability
Cohere API
Embeddings, reranking, and text generation in one — a top pick for RAG engineers
DeepInfra
Low-latency inference for open-source models, extremely transparent low pricing, supports crypto payment, includes image-generation models
EasyRouter
From Fu Sheng, 15% off storewide, 40+ model providers, DeepSeek discounts up to 75% off
HuggingFace Inference API
One-click access to hundreds of thousands of open-source models, the go-to for the ML community
hvoy.ai
A relay-station directory and review aggregator — not a relay itself, it helps you find the right one
Lepton AI
Former ex-Meta open-model inference platform, now part of NVIDIA DGX Cloud (lepton.ai redirects to NVIDIA)
01.AI (Lingyi Wanwu)
Founded by Kai-Fu Lee, dual open/closed-source Yi series, standout multilingual capability
Martian
Former commercial LLM-routing pioneer; the site is now Thesean AI, a research lab for best-execution LLMs
Mistral AI API
Europe's leading open-source AI player, GDPR-compliant, low-priced Codestral code generation
Modal
On-demand serverless GPU, minimal-friction deployment for custom model inference
ModelScope
Alibaba's open-source model community, with a free inference tier for Qwen/DeepSeek
Not Diamond
The most powerful intelligent model router for coding agents — 20%+ cost savings at frontier quality
Novita AI
200+ open-source models, dedicated image generation, low-cost global inference
PackyAPI
An early domestic relay optimized specifically for Claude Code, an active community, a $1 new-user bonus, a top pick for coding workflows
Perplexity API
Search-augmented AI, real-time web retrieval, built-in citations
PoloAPI
Balances stability with multi-model coverage, built around "set it up once and let it run," with enterprise governance features like usage stats and auditing
Replicate
Open-source model aggregation + custom model deployment, a leading platform for image/video generation, supports publishing private models
RunPod
Per-second billed GPU cloud, community compute pool, 70% cheaper than AWS
Segmind
Flux/SD image generation from as low as $0.001/image, a top pick for high-volume creative generation
Unify AI
No-setup AI agent teammates, accessing all models from 8 providers incl. OpenAI/Google/DeepSeek/Anthropic
Unity2.ai
30 billion+ tokens processed daily, a top pick for high-concurrency enterprise relay
Vidu
Shengshu Technology's Vidu video-generation model, one of the first domestic Sora-class video models with an official open API
WaveSpeedAI
1000+ image/video generation models accelerated in one platform, sub-second output, Flux/Seedance/Kling all in one place
Anyscale
From the team behind the Ray framework, enterprise-grade high-concurrency deployment and fine-tuning for open-source models
Eden AI
Single API aggregating OpenAI/Anthropic/Google, with built-in fallback routing
FlintAPI
Smart-routing SaaS — subscription + usage overage, 30+ Chinese LLMs under one API, $5 free credit for new users
FlowBar
Dual coverage of domestic + global models, 80+ model routes, 50,000 trial tokens for new users, USD pay-as-you-go, production-grade reliability
Inworld Router
220+ models with 0% markup, intelligent conditional routing, optimized for coding agents and AI companions
Privnode
$10 signup credit, deep Claude Code integration, focused on coding models
ProAI API
Rated by multiple communities as the most balanced, trustworthy choice — solves connectivity problems while keeping costs in check, with three-way Claude/GPT/Gemini coverage
RightCode
Coding-focused, top up from ¥1, clear documentation, Sonnet as low as ¥0.9/M
RunAPI
A mainland OpenRouter alternative — 300+ models with smart routing, Claude Code/OpenClaw support, runapi.host for mainland direct connect
AIFast.club
444+ models with mainland direct connect, recharge volume discounts (up to 7.5% off), a real-time status monitoring dashboard, unified single-key management
AiHubMix
Developer-friendly integration, permanent free testing tier, Prompt Caching support, solid documentation
Atlas Cloud
Image/video generation aggregator (Seedance/MiniMax/GPT Image etc.), 300+ models, transparent per-second billing
AzAPI
Multimodal creative-model aggregator: MJ/Suno/Luma/Kling/Flux/Udio; ICP-filed .com.cn domain; multi-group FX (Azure ¥0.4/USD)
Baidu Qianfan
Official Baidu platform, ERNIE model family, enterprise-grade SLA, supports model fine-tuning
ByteCat
Built for code engineers — full Codex Linux/Windows setup guides, mainland direct connect, AI coding model specialist
CloseAI
Self-described largest enterprise-grade AI relay platform in Asia, native support for the OpenAI/Claude/Gemini protocols, DPA data agreements available
JiekouAI
Full model coverage, subscription plans + pay-as-you-go, dedicated Cloud Code product, 30-second integration
LinkAi
Clear documentation, high bonus ratio on large top-ups — ¥100 gets you ¥30 free — covers both Claude and GPT
MegaLLM
70+ models aggregated through official channels, stable and highly available, a global AI API relay for mid-to-high-end needs
MoleAPI
Single API unifying GPT-4o/Claude/Gemini, <50ms direct-connect latency, OpenAI-format compatible, free credit for new signups
OAIPro
Official-channel relay, priced at parity with official rates, high stability
ofox.ai
Unified LLM gateway covering 100+ models including GPT/Claude/DeepSeek, built around Claude compatibility and undegraded official direct-source access
OpenClaw
An open-source personal AI assistant — free, self-hosted, can reuse subscription/gateway/local models
Huawei Cloud Pangu
Huawei's in-house large model, a top choice for high-compliance finance/government/healthcare scenarios
Relaydance
Pivoted to video-generation relay: one API for Seedance 2.0/2.5, plus Claude/Grok text and image models
UnoRouter
Multi-model aggregation, 300+ models (44+ upstreams), 253 free models, subscriptions with 2x credit
WinToken
Generous new-site bonus, roughly ¥113 in new-user trial credit, dual subscription/pay-as-you-go modes, broad model coverage
XycAi (Xingdao Intelligence)
Compliance-certified + Kling AI, a domestic/international mixed-model enterprise relay
YKH.AI
Minimalist pricing, Lite multiplier 0.2, performance-first positioning
Getimg.ai
All-in-one image generation + editing SaaS, bootstrapped and profitable for years, 3.5M+ monthly visits, creator-friendly
MiniMax Open Platform
MiniMax's official open platform — text, image, video, speech and music all behind one API
AI21 Labs API
Jamba hybrid architecture, low-cost long context, enterprise NLP specialization
AICloud Feiyun
50 free Sonnet calls given away daily, low-price direct connect to Claude
35.AIGCBEST
OpenAI-focused relay (now rebranded 'Qianduoduo API'), Azure pricing structure, exchange rate around 1.5¥/USD, stable and reliable
AIMLAPI
1000+ models via one endpoint — the go-to multi-model aggregator for overseas developers
ChatFire
Low-price multimodal aggregation, ¥0.5-1/USD exchange rate, a mix of domestic and international models, includes image and video generation
Claude API
Focused Claude relay, launched April 2026, supports image generation, ¥1 trial credit, competitively priced Opus access
DigitalOcean Gradient
An AI inference platform for the DigitalOcean developer ecosystem, pay-as-you-go open-source models
DMXAPI
480+ model multimodal aggregator, one-stop coverage of text/image/video, enterprise-grade stability
DuckCoding
$1 signup credit, tiered pricing, corporate invoicing supported, focused on coding use cases
Tencent Cloud Hunyuan
Tencent's official Hunyuan model, deeply integrated with the Tencent Cloud ecosystem
IKunCode
A coding-focused relay with an active QQ community, pure pay-as-you-go with no plans, covers Claude Fable 5/GPT-5.6 and other newest flagships
NodAPI
Fast multi-model aggregation, 545+ models, low-latency mainland direct connect, Alipay/WeChat/Stripe
Poixe AI
Operating since 2024, 17+ providers / 8 model families covered, free models + level-based discounts
UU API
MAX account pool, image generation at ¥0.04/image, ¥1 new-user bonus
iFlytek Spark
iFlytek's official platform — strong on both speech and text, the new Astron MaaS platform, vertical strength in education and healthcare
ModelsLab
One API replacing 10+ image/video/speech vendors, 10,000+ community models, claimed 99.9% uptime
302.AI
Pay-as-you-go with zero monthly fee — an enterprise-grade AI resource platform offering "one-stop access to every mainstream model," with balances that never expire
B.AI
Full model lineup + USDT/crypto payment, privacy-friendly, no real-identity binding required
Bob API
Solo-developer style, high-quality mainland direct connect, mainstream models, good for personal projects
Glama AI Gateway
An AI gateway with transparent price comparison, quick multi-vendor switching for experiments
Quanzil API
Full-coverage aggregated relay, 454+ models, WeChat/Alipay/crypto payments, invoicing available
KoalaAPI
Specializes in integrating mainstream overseas models (Gemini/ChatGPT/Claude), claims 99.7%+ success rate for Claude 4.5
MNAPI
Cross-vendor quality monitoring, all-in-one aggregation, 239+ models (Claude/GPT/Gemini/Grok/image & video)
Poe API
From Quora, subscription-based multi-model API, quick access to mainstream AI capabilities
Sub2API
A subscription-pooling relay station, group-buy-style low-cost shared access to Claude Max capacity
UiUiAPI
300+ large-model aggregation, enterprise-grade high-concurrency design, claims official-channel sourcing with transparent, quantified discount rates
XJAI (Hunter API)
Now rebranded 'Hunter API', multi-model aggregation, exchange rate around 1¥/USD, 533+ model coverage
Yinhe API
$0.4 signup bonus, dedicated Claude Code optimization, a good pick for lightweight testing
RunComfy
Cloud-hosted ComfyUI with one-click deploy, serverless workflow API, no GPU environment to set up
147API / 147AI
The "go-to pick" in multiple reviews — highly compatible interface, RMB settlement, enterprise-grade aggregation relay
AiGoCode
Subscription + pay-as-you-go; ¥50=$50 credit exchange (1:1); mainstream model coverage; try before you subscribe
Boluotu AI
⚠️ Announced pivot to enterprise-only: personal top-ups and new registrations closed; 1044+ models incl. newest flagships
Chutes
Low-latency hosting of SOTA open-source models (Kimi/GLM/DeepSeek/Qwen etc.); pay-as-you-go + subscriptions; unified OpenAI-compatible API
GGWK1
OpenAI/Claude/Gemini three-family aggregation, ¥0.6-1/USD exchange rate, mainland direct connect
Lumin AI
Launched in 2026; original luminai.cn has moved to luminai.cc, and the main domain now requires a proxy
MKEAI
Low-latency mainland direct connect to Claude/GPT/Gemini/Grok/DeepSeek, fast roll-out of new models (Grok 4.6/Gemini 3.7/Opus 5), new-user registration currently closed
No.1-API
A one-stop model-aggregation relay with well-documented interfaces, a unified entry point for Claude/GPT/Gemini/DeepSeek
StepFun
StepFun's official platform, Step-3.x series, multimodal flagship plus voice and image, low latency
TokenMix
Multi-model aggregation, flexible payment options, serves developers at home and abroad
V-API
Multi-model direct-connect relay covering Claude/GPT/Gemini/DeepSeek/Grok, stable mainland-China direct connect
YunWu API
500+ aggregated models, ¥0.5/USD exchange rate, free daily GPT-4o calls via GitHub login
Zhihui API
A new 2026 platform covering the GPT-5.6 / Claude Opus / Gemini / Grok families, with tiered recharge bonuses for members
Boxying
Ultra-low prices: Claude at 0.3x official, Codex at 0.21x, ¥1 = $1 FX; lightweight domestic relay
DawCode
A new 2026 platform, daily check-in rewards, ¥4 new-user trial credit
Nio API
Brand upgraded to newapi.gs, wildcard subdomains to resist blocking, currently GPT/Gemini-focused, legacy API stays compatible
Yiye Zhiqiu API
No minimum top-up, lightweight and test-friendly, zero-barrier proof of concept
UniAPI
OpenAI/Claude/Gemini three-way coverage + Midjourney/Suno/Kling and other multimodal models, mainland direct connect, channel-discount billing
Shenma Relay API
2557 multimodal models covered (text/image/audio/video), claims to be the most in the industry, pay-as-you-go with mainland direct connect
GPTGOD
A reverse-engineered ultra-low-price relay, exchange rate around ¥0.6/USD — extremely cheap but stability is not guaranteed
AIAPIpk
Discontinued (verified 2026-08-16): the domain no longer resolves; the comparison platform is unreachable
Baichuan API
Discontinued (verified 2026-08-16): the domain no longer resolves; service unreachable
Banana AI
Sunset: the serverless GPU inference platform was shut down on 2024-03-31
Chien API
Discontinued (verified 2026-08-16): the domain no longer resolves; service unreachable
Cooper-API
Discontinued (verified 2026-08-16): the domain no longer resolves; service unreachable
GPTAPI.US
Discontinued (verified 2026-08-16): the domain is parked; service unavailable
lxg2it ModelRouter
Discontinued (verified 2026-08-16): the domain no longer resolves; service unreachable
Meshs One
Discontinued (verified 2026-08-16): site unreachable; service unavailable
NanoBanana
Pivoted away from relay service (verified 2026-08-16): now a free Nano Banana image tool site
NativeAI API
Pivoted away from relay service (verified 2026-08-16): now the site of a Japanese software-development company
OAIPlus
Discontinued (verified 2026-08-16): site unreachable; service unavailable
PaintBot
Discontinued (verified 2026-08-16): site unreachable; service unavailable
SBGPT
Discontinued (verified 2026-08-16): the domain is for sale; service has shut down
Sulian AI
Discontinued (verified 2026-08-16): origin server unresponsive; service unavailable
TomCat API
Discontinued (verified 2026-08-16): the domain no longer resolves; service unreachable
Xingtu API
Redirected to an unrelated site; no longer a relay service (verified 2026-08-16)
ZHTec API
Discontinued (verified 2026-08-16): site unreachable; service unavailable
ShiyunApi ⚠️ Discontinued → TokenRiver
Discontinued; successor brand is TokenRiver 词元之河 (tokenriver.cn), now focused on domestic Chinese models
FAQ
What is an AI API relay, and why do mainland China developers use one?
An AI API relay (also called an API reseller) buys official API capacity from OpenAI, Anthropic, Google and others in bulk, then resells access at a lower barrier to entry through an OpenAI-compatible endpoint. For mainland China users, relays solve three problems at once: signing up without a card tied to an unsupported region, paying in RMB via Alipay/WeChat Pay, and reaching the model through a mainland direct-connect node instead of a VPN.
How is a relay different from the official API?
Technically, a relay exposes the same OpenAI-compatible interface as the official API — you only need to point base_url at the relay and keep the rest of your code unchanged. The differences are in price (often 10-40% below official pricing, though not always), reliability (depends entirely on the relay's own operations), and model fidelity (some relays quietly substitute a cheaper model for the one you requested). Picking a trustworthy relay is the whole game.
How do I tell if a relay is trustworthy?
Five things worth checking: (1) a real operating entity behind the site, not an anonymous side project; (2) the exact model versions are stated clearly, not vague marketing language; (3) independent, third-party usage reports exist somewhere online; (4) pricing is plausible — anything far below what the official API costs to run is a red flag for quietly-downgraded models; (5) small top-ups are supported, so you can test with a small amount before committing a large balance.
Why isn't every provider reviewed here in English?
We track 153+ relay providers on the Chinese side of the site, and English coverage is now essentially complete — every review here has been independently written and verified, the same as its Chinese counterpart. A small handful of the newest additions haven't been translated yet; if the provider you're looking for isn't listed here, the full Chinese-language comparison board always covers all of them.