aicodemirror lists 2 models: GPT-4o and Claude 3.5 Sonnet. For a relay station that explicitly targets developers using Claude Code, this is a hyper-specialized setup, not a general-purpose hub.
Model-by-Model Breakdown
GPT-4o
Context window: 100,000 tokens. This is the standard for GPT-4o across most relay stations, but it’s worth confirming because some cheaper relays cap it lower. Latency: 900ms average. That’s on the slower side for GPT-4o. Most direct OpenAI API calls land around 300-500ms. The extra 400ms likely comes from the relay’s routing layer. For interactive coding tasks (autocomplete, inline edits), you’ll feel the delay. For batch processing or code review, it’s acceptable. Stability: Uptime is 94.0%. This is mediocre. Over a 30-day month, you lose ~1.8 days of service. For a tool you depend on during work hours, that’s a gap. GPT-4o specifically may have higher availability, but the platform-level number is what we have. Speed rating: 3.5/5. The platform’s own score confirms it: this is not a speed demon. The 900ms latency drags it down. Bottom line: GPT-4o works here, but you’re paying a latency tax. If you need GPT-4o and can’t use OpenAI directly (VPN issue), it’s functional. If speed matters, look elsewhere.
Claude 3.5 Sonnet
Context window: Same 100,000 tokens. Claude models typically offer 200K on Anthropic’s own API. The 100K cap here is a real limitation for large codebase analysis or long conversation histories. Latency: Same 900ms average. Claude 3.5 Sonnet is usually faster on Anthropic’s endpoint (around 200-400ms). The relay adds significant overhead. Stability: Covered by the platform’s 94.0% uptime. No per-model SLA is published. During an incident, you won’t know which model fails first. This is a common gap in relay stations, and aicodemirror doesn’t fill it. Speed rating: 3.5/5. Consistent with the latency data. Key takeaways:
- Both models share the same 100K context window and 900ms latency
- No per-model availability SLA — you can’t predict which model fails first during an outage
- Claude 3.5 Sonnet’s context is halved compared to Anthropic’s native 200K
Pricing
| Model | Price Context | Notes |
|---|---|---|
| GPT-4o | Not explicitly listed per-token | Rates change; check dashboard for current |
| Claude 3.5 Sonnet | Not explicitly listed per-token | Rates change; check dashboard for current |
| Minimum recharge | Not specified | No minimum barrier to start |
| Payment methods | Alipay, WeChat Pay | Chinese developers: no foreign card needed |
| No promo code is available. The platform’s “competitive pricing” claim in its pros is unverifiable from the data — you’ll need to compare against other relay stations at the time of purchase. | ||
| Key takeaways: |
- Pay with Alipay or WeChat Pay — no VPN or foreign credit card required
- No minimum recharge means you can start small
- Rates change; the 2026 context means don’t assume current prices hold
Pros & Cons
Pros
- Code-focused relay: Designed for Claude Code users. The model selection reflects this.
- Good Claude support: Claude 3.5 Sonnet is the primary draw.
- Active updates: The platform receives ongoing development.
- Chinese payment methods: Alipay and WeChat Pay remove the payment barrier.
Cons
- Limited model selection: Only 2 models. If you need Gemini, DeepSeek, or any other model, this station won’t help.
- Mediocre uptime: 94.0% is below what most developers expect for a paid API relay.
- High latency: 900ms average adds perceptible delay to every API call.
- Newer station: Limited history means less track record for long-term reliability.
- Coding-only focus: Not useful for general-purpose LLM access.
Verdict
aicodemirror is a niche relay: it exists for developers who want Claude 3.5 Sonnet and GPT-4o without a VPN, and who are specifically using Claude Code. The model count is tiny (2), the latency is high (900ms), and the uptime is shaky (94.0%). The 100K context window is standard for GPT-4o but a downgrade for Claude. If you need Claude 3.5 Sonnet for coding and can’t access Anthropic directly, this works. If you need variety, speed, or reliability, skip it. The payment methods (Alipay, WeChat Pay) are the main convenience for Chinese developers. Bottom line: Use for Claude Code access with Chinese payment methods. Don’t use for anything else.
FAQ
Q: Can I use aicodemirror for non-coding tasks like chat or content generation? A: Technically yes, but the platform is optimized for coding use cases. The model selection (only GPT-4o and Claude 3.5 Sonnet) is limited for general-purpose work. You’d get better variety from a general relay station. Q: What happens when the 94.0% uptime fails — which model goes down first? A: The platform does not publish per-model availability SLAs. During an outage, you won’t know whether GPT-4o or Claude 3.5 Sonnet fails first. This is a common gap in relay stations. Q: Is the 100,000 token context window shared across both models? A: Yes, both GPT-4o and Claude 3.5 Sonnet are capped at 100,000 tokens on this relay. Note that Claude 3.5 Sonnet natively supports 200,000 tokens on Anthropic’s API — you lose half the context window here.
Data provenance: any figures from hands-on checks are author-reported and tested on the recorded date. Independent verification is unavailable unless a source is linked.
