GPT-4o input costs $0.00 per million tokens — because aicodemirror has no free tier (free_trial: False) and no published per‑token rates. The only way to pay is through Alipay or WeChat Pay, with no minimum recharge amount specified. That makes direct cost comparison impossible without a top‑up.
Pricing Structure
aicodemirror does not publish per‑million‑token pricing for either GPT‑4o or Claude 3.5 Sonnet. The platform uses a prepaid balance model — you recharge via Alipay or WeChat Pay, and each request deducts from your balance. No batch discounts, no volume tiers, no hidden per‑request overhead fees are documented. Bottom line: Without per‑token rates, you cannot compare aicodemirror’s pricing against direct API costs or other relay stations. The lack of a free trial (free_trial: False) means you must commit money upfront with zero visibility into unit economics.
Model Selection
Only two models are available: GPT‑4o and Claude 3.5 Sonnet. Both are code‑friendly, which matches the platform’s “Claude Code users wanting variety” description. No Gemini, no DeepSeek, no smaller or cheaper models. If you need a fallback for non‑coding tasks, you will need a second relay.
| Feature | aicodemirror | OpenRouter |
|---|---|---|
| Models available | 2 (GPT‑4o, Claude 3.5 Sonnet) | 200+ |
| Per‑token pricing | Not published | Published per model |
| Free tier | None (free_trial: False) | $1 credit on signup |
| Payment methods | Alipay, WeChat Pay | Credit card, crypto, Alipay |
| Min recharge (CNY) | Not specified | ¥10 |
| Uptime (data) | 94.0% | 99.9%+ (varies by model) |
| Latency (data) | 900ms | 500–1500ms (varies) |
Performance
aicodemirror reports 94.0% uptime and 900ms latency. The speed rating is 3.5/5, with a safety rating of 4/5. These numbers are below what most production‑grade relay stations deliver — OpenRouter and typical Chinese relays often sit at 99%+ uptime. The 900ms latency is acceptable for code completion but noticeable for chat. Key takeaways:
- 94.0% uptime means roughly 22 hours of downtime per month — risky for production.
- 900ms latency is workable for batch code review but slow for interactive chat.
- Safety rating of 4/5 suggests basic rate limiting and content filtering.
Pros & Cons
Pros
- Code‑focused: both models are strong at code generation and debugging.
- Active updates (noted in platform data).
- Alipay and WeChat Pay accepted — no credit card needed. Cons
- Only two models — no fallback for non‑coding tasks.
- No published per‑token pricing — cannot compare costs.
- No free trial — must recharge blindly.
- 94.0% uptime is low for a paid service.
- No batch discounts or volume pricing available.
Verdict
aicodemirror serves a narrow niche: Chinese developers who want Claude 3.5 Sonnet and GPT‑4o without a VPN and are comfortable prepaying without knowing per‑token costs. The 94.0% uptime and lack of pricing transparency make it a poor choice for production workloads. If you need model variety or cost predictability, look elsewhere. If you only need two code‑friendly models and can absorb the occasional downtime, it works.
FAQ
Q: Does aicodemirror offer a free trial?
A: No. The platform data explicitly lists free_trial: False. You must recharge via Alipay or WeChat Pay before making any requests.
Q: What are the per‑token costs for GPT‑4o and Claude 3.5 Sonnet?
A: aicodemirror does not publish per‑million‑token rates. The platform uses a prepaid balance model with no disclosed unit pricing. You cannot calculate cost per request without testing.
Q: Can I use aicodemirror for non‑coding tasks like creative writing or image generation?
A: No. The platform only offers GPT‑4o and Claude 3.5 Sonnet, both text‑only models. No image generation, no embedding models, no vision (unless the model supports it inherently). It is designed for code‑focused users.
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.
