Does Claude Work in China? How to Access Claude in 2026

Quick Answer

Does Claude work in China? No, not through Anthropic's official channels. Anthropic does not offer service in mainland China, claude.ai is blocked, and the official API at api.anthropic.com is unreachable from the mainland without a proxy or VPN. That said, Chinese developers in 2026 have several workable routes to use Claude, ranging from VPN-plus-direct access to third-party relay stations and API gateways like TeamoRouter. The methods differ dramatically in stability, cost, and security — this article ranks them so you can pick the right one.

Why Claude Is Not Officially Available in China

The reason is straightforward: Anthropic does not operate in mainland China and has not opened the region for service. The two official entry points are both blocked in practice:

• claude.ai — the Web app is not accessible from mainland IPs.
• api.anthropic.com — the API endpoint is unreachable without a proxy or VPN.

That means the phrase "Claude works in China" is only true with a workaround. Which workaround you choose determines how well it works, how much it costs, and how risky it is.

The Four Routes to Use Claude from China in 2026

Method 1: VPN + Direct Access

How It Works

You subscribe to a VPN, connect to a server in an Anthropic-supported region, and then use Claude normally — either in the browser at claude.ai or through the API at api.anthropic.com.

Pros

• Full official experience with no intermediaries.
• Data flows directly to Anthropic (through the VPN tunnel).
• Official prompt caching works as designed.

Cons

• The VPN must stay up for the entire session; a drop cuts off Claude mid-task.
• Added latency: China → VPN server → Anthropic adds delay on every request.
• VPN protocols are periodically interfered with on mainland networks.
• Persistent VPN connections can flag accounts for anomalous-login review.
• You still pay for a VPN on top of Claude API or subscription costs.

When to use it: occasional short Claude Web sessions. Not recommended for daily coding or long Claude Code sessions, where a dropped tunnel is painful.

Method 2: Mirror Sites — High Risk, Avoid

Mirror sites are third-party clones of the Claude Web interface that proxy requests to Anthropic's API behind the scenes. They are often free or suspiciously cheap.

Why You Should Not Rely on Them

• You cannot verify what happens to your prompts. Free mirrors have no business model other than your data or your traffic.
• They usually break prompt caching, which causes token costs to balloon when you pass through them.
• They frequently go down and can disappear overnight with your conversation history.
• They do not support agentic tools properly, so Claude Code and similar tool-using workflows fail.

If a deal looks too good to be true, it is. Mirror sites are a leaky and insecure way to reach Claude, and they are not a serious option for production work.

Method 3: Relay Stations (中转站)

Relay stations, also called "relay" or "中转" services, provide a shared API key and a re-sold endpoint that forwards requests to Anthropic. They are popular because they are easy to use and do not require a VPN.

Pros

• No VPN needed; they handle connectivity.
• Cheap entry, often pay-per-use.
• Easy to switch between models.

Cons

• Prompt caching is frequently broken, and since Anthropic bills cache reads at a steep discount, a relay that misses the cache silently multiplies your effective cost.
• Throughput and rate limits are shared across all customers, so latency and quota are unpredictable.
• You depend on the relay's financial health — many shut down without warning.
• Less control over which region your requests actually land in.

When to use it: casual chat where cost predictability is not critical. Be wary for Claude Code, agents, or anything that makes many sequential tool calls.

Method 4: API Gateway — The Recommended Route

An API gateway such as TeamoRouter is a different category from a relay. It gives you a stable, globally reachable API endpoint that works from mainland China without a VPN, while preserving official prompt caching and supporting agentic tool use fully.

Why an API Gateway Is Different

• Stable connectivity: the endpoint is reachable from China directly, so no VPN tunnel can drop mid-session.
• Prompt caching preserved: gateway routing keeps Anthropic's cache behavior intact, which matters because cache hits cost a fraction of cache misses.
• Full agent compatibility: Claude Code and other tool-using clients work without special hacks.
• Single API Key: one key covers Claude, GPT, Gemini, and DeepSeek, so you are not juggling providers.
• Transparent billing: pay-as-you-go with a dashboard, rather than opaque shared accounts.

How to Use Claude from China with Claude Code

The fastest way to use Claude from China in 2026 is to point an official Anthropic client at a gateway. Here is how that looks with Claude Code:

# Route Claude Code through an API gateway
export ANTHROPIC_BASE_URL=https://api.teamorouter.com/anthropic
export ANTHROPIC_AUTH_TOKEN=$TEAMO_API_KEY
export ANTHROPIC_MODEL=claude-sonnet-4-5

claude

With the environment variables set, Claude Code connects to the gateway endpoint instead of api.anthropic.com, which means no VPN and no proxy configuration are needed. The same approach works with the Anthropic SDKs:

import anthropic

client = anthropic.Anthropic(
    base_url="https://api.teamorouter.com/anthropic",
    api_key="your-teamorouter-key",
)

resp = client.messages.create(
    model="claude-sonnet-4-5",
    max_tokens=1024,
    messages=[{"role": "user", "content": "Explain how prompt caching works."}],
)
print(resp.content[0].text)

What "Base URL" Means and Why It Matters

Anthropic SDKs let you override the endpoint with base_url. When you point that at a gateway, you keep using the official SDK and official models, but the connection is routed through the gateway. This is the same mechanism Chinese developers use to keep Claude working reliably without maintaining their own proxy infrastructure.

Cost Comparison and What to Watch

• VPN route: VPN subscription + full official API prices. Predictable but adds monthly VPN cost and setup overhead.
• Mirror sites: "free" or cheap, but you pay with data and reliability.
• Relay stations: cheap per-token, but broken caching can quietly multiply real cost.
• API gateway: pay-as-you-go at official-like rates, no VPN cost, cache benefits preserved.

The hidden variable is prompt caching. With official API pricing, cache reads are dramatically cheaper than full input tokens. A route that preserves caching (VPN-direct or a well-run gateway) will be far cheaper for agentic, multi-turn workloads than one that breaks caching (most mirrors and many relays).

Security and Compliance Notes

• Data residency: official API requests are processed by Anthropic in supported regions. Choose a route you can verify for sensitive or proprietary code.
• Account safety: logging in from VPN IPs repeatedly can trigger review. Using a gateway avoids that pattern.
• Don't paste secrets into mirrors: a mirror sees everything you type.

Common Setup Mistakes to Avoid

When setting up Claude access from China, a few mistakes cause most of the failures:

1. Pointing the client at the wrong base URL. The Anthropic SDK expects base_url to end in /anthropic for gateway routing. Omitting the path (or adding it twice) produces confusing 404s.
2. Mixing up API keys. A gateway key is different from an Anthropic platform key. If you set ANTHROPIC_AUTH_TOKEN to the wrong key, authentication fails in a way that looks like a network problem.
3. Relying on a VPN that drops. Browser sessions look fine until the tunnel drops mid-conversation. For long agent runs, a VPN is fragile; a gateway is not.
4. Assuming a relay preserves caching. Before committing to a cheap relay, run a two-turn conversation and check the token bill. If the second turn charges full input tokens, caching is broken and your costs will multiply.

Frequently Asked Questions

Can I get banned for using Claude from China?

Anthropic does not officially serve mainland China, so access relies on workarounds. Using a stable gateway is far less likely to trigger account reviews than repeatedly logging in from rotating VPN IPs. Always use credentials you control and avoid sharing accounts.

Is using a relay station illegal?

Using a relay is a gray area rather than a clearly defined legal issue in most contexts, but it is risky from a security and terms-of-service standpoint. This article does not constitute legal advice; the practical point is that relays are less stable and less transparent than a direct gateway.

Do I still need a VPN with TeamoRouter?

No. The entire point of the gateway route is that the endpoint is reachable from mainland China without a VPN. You can remove the VPN dependency for Claude traffic and keep it only for anything else you need it for.

Which Claude models can I use through a gateway?

Gateways typically expose the same model catalog as the official API, so the models available to you follow your account and billing tier. If a model is available on the official API, it is generally available through a gateway that routes to it.

Bottom Line

Does Claude work in China? Not through official channels, but yes through reliable workarounds. For serious work — daily coding, agents, production applications — the pragmatic 2026 answer is a stable API gateway like TeamoRouter: no VPN, no dropped tunnels, official caching, and full tool support. For occasional browser use, a VPN may be enough. Mirror sites and casual relays are best avoided for anything that matters. Start with the gateway route and you can have Claude Code running from China within minutes.