Awesome Claude API Relay Resources

中文版 | English

A carefully curated list of reliable Claude API relay services, helping developers quickly find stable and cost-effective Claude API providers. All information comes from official websites to ensure timeliness and accuracy.

📊 Monthly Traffic Overview (2026-09)

This edition compares monthly visits for 2026-08 vs 2026-09 across 22 channels (the shut-down Unity2.Ai has been removed; LinkAGI, Kunavo, TokenDos, AI SaaS Go, API Kuailian and AnyAIGC are newly added).

• Stable leaders: OpenRouter 22.0M → 21.3M (-3.4%), still the only ten-million-level channel; the top-tier landscape remains stable.
• Outsized gains: API Kuailian +7785.8%, Kunavo +2778.2%, LLM Model Mirror +310.1%, B.AI +78.6%, AIHubMix +12.1%. The top three are low-base new sites or tool sites.
• Broad declines: AIGoCode -58.6%, EasyRouter -50.6%, ZenMux -43.9%, TokenDos -29.2%, YesCode -16.0%, IKunCode -15.3%, CCTest -14.5%, TeamoRouter -13.0%, LinkAGI -10.0%, OfoxAI / Ofox Model Verifier -3.8%.
• Flat and no-data: CoderPlan and FerryAPI show 0 estimates in both months; AICoding dropped to 0 in September (August value carried over from the previous platform; -100% leaves no base to back-calculate); AI SaaS Go and AnyAIGC have no third-party estimates yet.
• Newly added: API Kuailian 162 → 12.8K (+7785.8%), a new site launched 2026-06 that started gaining traction in September; LinkAGI 10.6K → 9.6K (-10.0%), Kunavo 156 → 4.5K (large growth), TokenDos 4.9K → 3.5K (-29.2%) — all three launched in 2026; AnyAIGC has no third-party estimates yet (brand started 2026-04; migrated from anyaigc.com to the current domain in 2026-09).

📋 Table of Contents

Traffic Overview | Introduction | Usage Notes | API List | Detection Tools | Contributing | Disclaimer

🚀 Introduction

This project collects stable and reliable Claude API relay services worldwide, providing developers with an authoritative, frequently updated reference. All listed API services have passed preliminary screening for basic availability and stability.

📝 Usage Notes

1. All services are provided by third parties; read each provider's official documentation and terms of service carefully before use
2. Try free credits or a small plan first, and confirm the service fits your needs before committing to long-term payment
3. Prices and policies may change; always refer to the provider's official website for the latest information

🔌 API List

Provider	Pricing	Review	Tags
OpenRouter openrouter.ai Monthly visits: 22.0M (2026-08) → 21.3M (2026-09) ▼ Down 3.4%	• No subscription, pay-as-you-go; unified credits cover all model fees • Transparent pricing, different rates per Claude version • Free trial credits for new users	• World's largest LLM aggregation platform: 5M+ users, 70T tokens processed per month • Supports the full Claude lineup (incl. the latest Claude Opus 4.7) • Multi-upstream automatic failover, high availability • Fully OpenAI SDK-compatible, near-zero code changes • Edge-node acceleration, fine-grained data privacy controls, team management	World-renowned All Claude versions Highly reliable Great SDK compatibility
AIHubMix aihubmix.com Monthly visits: 253.6K (2026-08) → 284.3K (2026-09) ▲ Up 12.1%	• Pay-as-you-go with no forced subscription; prices roughly in line with official rates • 10% long-term discount for new users and partners • Transparent, auditable billing; multiple payment methods	• Leading global LLM aggregation platform with full Claude support (incl. latest Opus 4.7 and Sonnet 4.6) • Processes over 100B tokens monthly; used by a large number of enterprise developers, with extremely high stability • Automatic edge-node routing, fast global access, strong concurrency • Fully compatible with OpenAI and Anthropic official API protocols, zero code changes • Supports web search, multimodal, video generation and more; very fast model updates	Multi-model Highly stable Low latency Great SDK compatibility Fast model updates
ZenMux zenmux.ai Monthly visits: 678.4K (2026-08) → 380.9K (2026-09) ▼ Down 43.9%	• Both pay-as-you-go (PAYG) and subscription plans • Fully compatible with OpenAI and Anthropic official protocols, no extra adaptation cost • Transparent billing with detailed usage records, no hidden fees • Innovative AI anomaly compensation: automatic refunds when output quality falls short	• Top-tier global model aggregation platform; only offers official/authorized original models, no downgraded or defective versions • Multi-provider automatic failover plus Cloudflare global edge acceleration for very high stability • Built-in smart model routing that automatically matches the best value model • Multi-dimensional consumption dashboards for cost optimization • Used by many enterprises and developers worldwide; supports personal/team/enterprise needs	Internationally known Highly stable Multi-model Anomaly compensation Enterprise-grade
IKunCode api.ikuncode.cc Monthly visits: 81.8K (2026-08) → 69.3K (2026-09) ▼ Down 15.3%	• Pricing available on the official model marketplace page • Multiple billing modes supported	• Complete API documentation • Full-featured console, easy to manage • Suited for developers with some technical background	Good documentation Easy management
AIGoCode www.aigocode.com Monthly visits: 39.4K (2026-08) → 16.3K (2026-09) ▼ Down 58.6%	• Subscription plans:   - Standard: ¥399 / 4 weeks, $110 weekly quota, 10% off   - Premium: ¥899 / 4 weeks, $260 weekly quota, 14% off   - Professional: ¥1799 / 4 weeks, $530 weekly quota, 16% off   - Enterprise: ¥9999 / 4 weeks, customized service • Flexible credits: ¥50 buys $50, never expires, pay per use • Subscription quota is consumed first; falls back to flexible credits automatically	• 99.9% service stability, 5M+ tasks processed • 1-on-1 dedicated engineer support with fast response • Team collaboration with shared quotas and permission management • Suitable for both individual developers and enterprise teams • Transparent billing, no hidden fees	High availability Team collaboration Multiple plans Good customer support
YesCode co.yes.vg Monthly visits: 21.7K (2026-08) → 18.2K (2026-09) ▼ Down 16.0%	• Free tier: 1M tokens per month • Paid plans: see official pricing page • Unified billing across platforms	• Excellent performance, average latency of just 38ms • Automatic failover, high stability • Multi-model routing for Claude Code, Codex, Gemini and more • Team key management and real-time usage stats • SOC 2 Type II certified, high security	Highly stable Low latency Multi-model Team-friendly
AICoding aicoding.sh Monthly visits: 2.3K (2026-08) → 0 (2026-09) ▼ Down 100.0%	• New-user perk: $5 free credit for joining the community • Detailed pricing visible after login	• Seamless integration with OpenClaw, VSCode and other mainstream IDEs • Dynamically scheduled high-quality account pool with smart switching to avoid bans • One-click install, zero configuration, ready out of the box • 10,000+ teams on board, stable and reliable service	IDE-friendly Ban-resistant Easy deployment New-user perks
EasyRouter easyrouter.io Monthly visits: 9.2K (2026-08) → 4.6K (2026-09) ▼ Down 50.6%	• No subscription; credits offset compute usage, pay for what you use • Site-wide 15% off token consumption in May; DeepSeek-v4-Pro at 75% off • Multiple credit tiers: $20/$100/$500/$1500	• Built by Fu Sheng's team, strong backing • Unified OpenAI-compatible interface: one line of code to switch, zero migration cost • 40+ top models supported, covering the full Claude lineup • Multi-region global deployment with smart routing, P99 latency under 200ms • 99.9% availability SLA with automatic failover and load balancing	All models Low latency Highly stable Major backer Great SDK compatibility New-user discounts
B.AI b.ai Monthly visits: 716.6K (2026-08) → 1.3M (2026-09) ▲ Up 78.6%	• One API key covers Claude + GPT + Gemini + major Chinese models • Blockchain-address login with fully anonymous payments (email/Visa/Master/Apple Pay supported) • Direct official API connection, zero tampering, lowest prices around	• AI model aggregation platform launched by Justin Sun's team • Official announcement: Justin Sun's Twitter • BAIClaw has launched a "Justin Sun's Brain" feature for trading crypto with his mindset	Newly launched Justin Sun's team All models Anonymous payment Crypto-friendly
OfoxAI ofox.ai Monthly visits: 221.1K (2026-08) → 212.7K (2026-09) ▼ Down 3.8%	• No subscription, pay-as-you-go; unified credits cover all model fees • Transparent pricing, different rates per model • Free trial credits for new users • WeChat Pay, Alipay and credit cards accepted	• One API key for 100+ models (GPT-5.4, Claude Opus 4.6, Gemini 3.1, etc.) • Native support for OpenAI, Anthropic and Gemini SDK protocols, zero code changes • Hong Kong direct-connect lines: 100-300ms latency from mainland China, no proxy needed • Multi-upstream smart routing with a 99.9% availability SLA • Team collaboration, shared billing, API usage management and more	Mainland direct-connect Multi-model Low latency Great SDK compatibility Team-friendly
CoderPlan coderplan.ai Monthly visits: 0 (2026-08) → 0 (2026-09) Flat	• Pay-as-you-go with no monthly commitment; top-up tiers: ¥30/¥100/¥300/¥500 (bonus credit on large amounts) • Claude Opus 4.7 at ~12% of official price (88% off), GPT series at ~7% (93% off) • Free trial credit for new users • Alipay, WeChat Pay and other mainland payment methods	• Full Claude lineup (Opus 4.7/4.6, Sonnet 4.6) plus GPT, Gemini and other major models • Deep integration with Claude Code, Cursor, Codex CLI, Gemini CLI and other AI coding tools • Smart cache-reuse mechanism that significantly cuts repeated-call costs • Transparent billing: each call logs model, tokens, cache savings and cost • OpenAI API-compatible, zero code changes • Public system status page, observable service stability	Great prices Mainland payments Great SDK compatibility Multi-model Cache savings
FerryAPI www.ferryapi.io Monthly visits: 0 (2026-08) → 0 (2026-09) Flat	• Prepaid USD balance deducted by actual usage; no fixed monthly fee or subscription • $5 free credit for new users • 1:1 matching bonus for inviting your first paying user • Model pricing on the official pricing page	• Low-cost OpenAI-compatible AI API gateway; migrate existing OpenAI SDK apps with one line of code • Aggregates 9 model providers: Claude, OpenAI, Gemini, DeepSeek, Qwen, Zhipu AI, Kimi, MiMo and Doubao — all behind one API key • Provider account-pool mechanism with automatic multi-upstream scheduling • Suited for customer service, translation, content generation, summarization, classification and automation • Solid official docs: ferryapi.io/docs	Claude support Multi-model aggregation Low cost New-user perks Referral rewards
TeamoRouter teamorouter.com/zh Monthly visits: 265.5K (2026-08) → 231.1K (2026-09) ▼ Down 13.0%	• Current Claude prices (per 1M input/output tokens, per the official model page): Opus 5.5 $1.144 / $5.72 (~29% of official), Sonnet 5.5 $0.568 / $2.84 (28%), Fable 5.1 $3 / $15 (30%), Haiku 5.5 $0.0286 / $0.143 (~29%) • Cache billed at Anthropic's official ratios; per-request token usage and charges can be audited item by item in the console • Alipay top-ups, pay-as-you-go — no overseas bank card needed • Failed requests are not billed; no minimum top-up	• Claude models on offer: Opus 5.5 / Sonnet 5.5 / Fable 5.1 / Haiku 5.5 as the current mainstays, with older models (Opus 5, Sonnet 5, Opus 4.6/4.7/4.8, Sonnet 4.6, Haiku 4.5) still listed • Dual-protocol support: Anthropic Messages (Claude Code) + OpenAI /v1/responses (Codex) — one API key works for both; just change the base_url and existing code runs immediately • Pre-configured in CC Switch and Claude Code Router — no need to enter API key / base URL manually • Mainland direct-connect: no VPN required, multiple backup lines • Also aggregates GPT 6.x / 5.x, Gemini, DeepSeek, GLM, Grok, Kimi and more — 59 models in total • Official model page publishes context length, time-to-first-token, output speed and availability SLA per model (self-reported data) for comparison • Docs and tutorials: teamorouter.com/zh/blogs	Multi-model aggregation Dual-protocol From 28% of official price Cache at official ratios Mainland direct-connect Pay-as-you-go Claude Code support Codex support One-click setup CC Switch built-in Claude Code Router preset
LinkAGI api.linktoagi.com Monthly visits: 10.6K (2026-08) → 9.6K (2026-09) ▼ Down 10.0%	• RMB pay-as-you-go, no subscription, pay for what you use • GPT-5.6 three-tier account pools: Luna ¥0.15 / 1M input tokens (output ~¥0.90), Terra ¥0.375 (output ~¥2.25), Sol ¥0.75 (output ~¥4.50) • ¥0.2 trial credit for new users • WeChat Pay and Alipay accepted; per-call token usage and charges auditable	• Unified API access for Codex, Claude Code, Gemini CLI and OpenAI-compatible clients • Supports OpenAI / Claude / Gemini protocols — root URL for Claude Code, /v1 for Codex, auto-composed per client • First-hand GPT account pools with a live model marketplace showing model names, groups and prices • Per-call logs record model, tokens, response status and actual charges • Free tools included: endpoint diagnostic, error troubleshooting center (401/403/404/429/504), cost calculator, open-source self-check script • Thorough Chinese onboarding docs, mainland direct-connect, clear human-support channels	RMB billing Pay-as-you-go Claude Code support Codex support Multi-protocol Mainland direct-connect Transparent pricing
Kunavo kunavo.com/zh Monthly visits: 156 (2026-08) → 4.5K (2026-09) ▲ Up 2778.2%	• Prepaid wallet: $10 minimum top-up, no subscription, no monthly fee, balance never expires • Claude prices (per 1M input/output tokens): Sonnet 5 $1.40/$7.00, Opus 5.5 $2.80/$14.00, Opus 5 $3.50/$17.50, Fable 5.1 $7.00/$35.00, Sonnet 4.6 $2.10/$10.50, Haiku 4.5 $0.70/$3.50 • Top-up bonuses: $10 on $100, $200 on $1000, $1250 on $5000 • Native Stripe billing: credit cards, Apple Pay, Google Pay, Alipay, WeChat Pay, plus local methods like Brazil Pix, Korea KakaoPay and India UPI	• OpenAI-compatible gateway aggregating frontier models from OpenAI, Anthropic and Google — 35 online models, most below official prices with savings up to 70% • Migrate by changing one line of base_url; drop-in OpenAI SDK replacement with full protocol compatibility for streaming / function calling / tool use / vision • Automatic failover: each request falls back across up to three upstream channels in order; recently failed channels are bypassed before the request is even sent • Native streaming SSE pass-through — time-to-first-token matches upstream official; no buffering, no batching • Prompt cache hits billed at 10% of input price (5% for Claude Opus 5.5, 2.5% for Fable 5.1), saving up to 90% • Fine-grained usage analytics by model / API key / IP, webhook pushes, CSV export, monthly caps per key and IP whitelists • 53 client integrations (Claude Code, Cline, OpenClaw, n8n, etc.), each verified against official docs with dates noted • Transparent pricing with no hidden multipliers; failed requests not billed; US East Anycast edge access	OpenAI-compatible Multi-model aggregation Up to 70% off Automatic failover Prompt caching Stripe billing Full Claude lineup
TokenDos www.tokendos.com/tokendos Monthly visits: 4.9K (2026-08) → 3.5K (2026-09) ▼ Down 29.2%	• Real-time per-token billing, no subscription • Shared marketplace pricing: aggregates idle API keys from providers, significantly cheaper than direct official access • Claude Sonnet 5: $0.022 / 1M input tokens, $0.11 / 1M output tokens • Prompt cache priced separately: cache read $0.0022, cache write $0.0275 (per 1M tokens) • GPT-6 Luna as low as $0.000744 / 1M input tokens, $0.003722 / 1M output	• AI API shared marketplace: connects providers with idle API keys to demand, monetizing idle compute while cutting on-demand costs • Dual-protocol endpoints, drop-in replacement for existing code • Model coverage: GPT-6 series (astra / sol / luna), GPT-5.6 (sol / terra), GPT-5.5, Grok 4.6 / 4.7 and Claude Sonnet 5 • Separate cache read/write pricing per model, cheaper for repeated long-context reads • New site (launched 2026-06); payment methods and team features not yet published on official pages	Shared marketplace Pay-as-you-go Ultra low cost Dual-protocol Prompt caching Claude support
AI SaaS Go aisaasgo.org/zh Monthly visits: No third-party estimate available	• RMB credited 1:1 by amount paid (¥1 = $1 credit); ¥5–¥1000 top-ups, no bonuses • Claude prices (per 1M input/output tokens): Sonnet 5 / 4.6 / 4.5 all $0.9/$4.5, Opus 5 $1.5/$7.5, Opus 5.5 $2/$10, Fable 5 / 5.1 $3/$15 • Precise separate metering for input, output, cache write and cache read • Bank cards and Alipay accepted • GPT models trigger long-context billing on single inputs over 272K tokens (input and cache read at 2x, output at 1.5x) • Video / audio / image models billed separately; failed tasks not charged	• One interface for OpenAI, Claude, Gemini, Grok, Kimi, DeepSeek and more • Claude coverage of 11 models: Fable 5 / 5.1, Opus 5.5 / 5 / 4.8 / 4.7 / 4.6 / 4.5, Sonnet 5 / 4.6 / 4.5 • Compatible with the OpenAI Responses API (native SSE streaming) plus the native Claude interface (x-api-key auth supported) • Multimodal coverage: text, images (GPT Image, Midjourney), video (Seedance), audio (Suno, Doubao Seed Audio) • Console shows API keys, call logs, token usage and actual spending • Complete error-code docs (401/402/429/5xx mapped to key, balance, rate limit, temporary failure) • Note: third-party relay reviews have recorded below-average availability; validate with a small amount before production use	Full Claude lineup Multimodal Transparent pricing RMB top-ups Dual-protocol
API Kuailian apikl.ai Monthly visits: 162 (2026-08) → 12.8K (2026-09) ▲ Up 7785.8%	• Prepaid balance, pay-as-you-go; no monthly fee, balance never expires; top up via Alipay / WeChat Pay / cards (Stripe); small top-ups and invoicing supported • Pricing model: official USD prices converted at ¥1 = $1, then multiplied by the group rate; input / output / cache write / cache read metered separately • Claude-Max (flagship) at 0.8x: Sonnet 5 ¥1.6 / ¥8, Opus 5 ¥4 / ¥20, Fable 5 ¥8 / ¥40 (per 1M input/output tokens) • Claude Code (AWS, official channel) at 3.8x: Sonnet 5 ~¥7.6 / ¥38, Opus 5 ~¥19 / ¥95 • Cache read at 0.1x input price, cache write at 1.25x, 1-hour cache write at 2x	• All 41 groups fully priced, with upstream source in the group name: Claude-Max is a Claude Max subscription pool, Claude-Plus is a Kiro channel (not official direct), Claude Code (AWS) is the official AWS channel, and degraded ("distilled") groups are clearly labeled — pick the AWS group for genuine models • Claude coverage of 14 models (19 model IDs incl. date-suffixed aliases): Fable 5 / 5.1, Opus 4.5 / 4.6 / 4.7 / 4.8 / 5 / 5.5, Sonnet 4.5 / 4.6 / 5 / 5.5, Haiku 4.5 / 5.5 • Three-protocol support: Anthropic /v1/messages, OpenAI-compatible, and native Gemini; official tutorials for Claude Code / Codex / CC Switch / Gemini CLI, plus a CDN-free direct line at dc.apikl.ai • Independent third-party testing (Veridrop, 149 runs from 2026-07 to 10) scored a median 95/100 "excellent" with no site-wide severe issues across 149 runs; however, the OpenAI-side token billing pass rate was only 34% and Claude thinking signature 72.4% — self-test if you are cost-sensitive or need strict authenticity • Launched 2026-06, domain registered 2026-07; customer support via email + QQ + WeChat + Telegram, with public refund rules and invoicing policy	Fully priced groups Dual-protocol Claude Code support Third-party tested RMB top-ups Official channel option New site
AnyAIGC anyaigc.ai Monthly visits: No third-party estimate available	• Pay-as-you-go, no subscription; pricing API /api/pricing publicly accessible (no login) with 333 models incl. 16 Claude models and all group rates listed • Current Claude prices (per 1M input/output tokens, per the model marketplace): Claude-Code-1 group ~24% of official — Sonnet 5.5 $0.476 / $2.38, Opus 5.5 $0.953 / $4.76, Fable 5.1 $2.38 / $11.9; Kiro-Claude-1 group ~12% — Sonnet 5.5 $0.238 / $1.19 • Huge price gaps between groups — picking wrong costs you: AWS-Bedrock-1 ~60% of official, AWS-Bedrock-2 ~80%, while Anthropic-Claude-1 and AWS-Claude-3 run at 1.49x (49% above official) • Cache read 0.05x (Sonnet / Opus), 0.025x (Fable), 0.1x (Haiku); cache write 5m at 1.25x, 1h at 2x • WeChat, Alipay (two channels) and cards accepted; $1 minimum top-up, RMB settled at 6.718 rate, invoicing available • No free credits (check-in disabled, no signup bonus); 5% referral rewards only	• Claude coverage of 16 models: Opus 5.5 / 5 / 4.8 / 4.7 / 4.6 / 4.5 / 4.1 / 4, Sonnet 5.5 / 5 / 4.6 / 4.5, Fable 5.1 / 5, Haiku 5.5 / 4.5 • Upstream channel in each group name: AWS Bedrock, AWS Claude, Anthropic, Azure, Vertex, Claude-Code, Kiro — all official or subscription channels; no "reverse-proxy" groups observed • Dual protocol: Anthropic /v1/messages + OpenAI /v1/chat/completions; change base_url and you're set; uptime status page and onboarding docs available • Brand operating continuously since 2026-04-22: the old site anyaigc.com stopped signups on 2026-08-09 and top-ups on 2026-09-10, with official notices directing users to anyaigc.ai; the migration was due to an upgrade to a USD-denominated billing system (the new site prices everything in USD, settled in RMB at a 6.718 rate) • Active operations: 394 site notices (latest 2026-10-22), claude-opus-5-5 cumulative usage of 22.4B tokens • Unverified items: operating entity undisclosed (whois fully privacy-protected), no third-party test scores yet (Veridrop lists it but no snapshot); validate with a small amount in cost-sensitive scenarios	Dual-protocol Full Claude lineup Fully priced groups Cache billing WeChat & Alipay $1 minimum Invoicing

🛠️ Detection Tools

Tool	Description	Features
CCTest cctest.ai/zh Monthly visits: 42.6K (2026-08) → 36.5K (2026-09) ▼ Down 14.5%	A professional Claude API detection platform that checks model authenticity and token billing fairness, trusted by dozens of leading relay providers	• Black-box detection mode — never stores user API keys, safe and reliable • Supports Claude Sonnet 4.6, Opus 4.6, Opus 4.7 and other mainstream versions • 11 rounds of professional tests to detect abnormal token multipliers and unreasonable charges • Transparent test cost: a single run consumes about $0.3
Ofox Model Verifier ofox.ai/verify Monthly visits: 221.1K (2026-08) → 212.7K (2026-09) ▼ Down 3.8%	Ofox's model verification tool: enter an API endpoint and verify in 15 seconds whether a relay's model is genuine	• Research shows 45.83% of relay sites engage in model substitution • Enter an API endpoint and verify the real model in 15 seconds • Provided by the OfoxAI team over Hong Kong direct-connect lines
LLM Model Mirror llmtest.cn Monthly visits: 834 (2026-08) → 3.4K (2026-09) ▲ Up 310.1%	One-click detection from multiple technical dimensions of whether a relay's Claude Opus/Sonnet/GPT/Gemini and other premium models are genuine — say no to wrapper scams	• Built on a GitHub open-source project with pure frontend logic • Zero-log privacy promise: API keys are burned after testing, nothing kept on servers • Multi-dimensional cross-validation: response protocol, response structure, knowledge Q&A, identity consistency, chain-of-thought traces, signature fingerprints • Authenticity score: 80%+ highly trusted / 50-79% questionable / below 50% high risk • Directly accessible from mainland China, no VPN needed

🤝 Contributing

PRs adding more quality Claude relay API resources are welcome. When submitting, please make sure:

1. All information is truthful and accurate, sourced from official channels
2. You include the official website link, pricing structure, an objective review, and data such as monthly visits
3. The service genuinely provides Claude API relay capability
4. No advertising or promotional misstatements

Before submitting, check whether the same provider is already listed to avoid duplicates.

⚠️ Disclaimer

This project only collects and organizes resources, and provides no warranty for the quality, security, or availability of any third-party service. Use third-party services at your own risk; the maintainers of this project assume no liability for any usage-related issues.

---

⭐ If this project helps you, please give it a Star!

⭐ Star History