GPT-4o input costs ¥1 per million tokens on TimiCC’s reverse proxy, and Claude 3.5 Sonnet output runs ¥5 per million tokens.
Free Tier & Registration
You get ¥1 free credit just for signing up. No credit card required — only a QQ email address. This is enough to run roughly 200 GPT-4o completions or 50 Claude Sonnet requests before you hit zero. Bottom line: ¥1 free credit covers basic testing; QQ-only registration is a privacy trade-off.
Per-Token Pricing Breakdown
TimiCC’s pricing model is split by model and direction (input vs output). The platform explicitly lists these rates:
| Model | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| GPT-4o | ¥1 | ¥2 |
| Claude 3.5 Sonnet | ¥1 | ¥5 |
| Compare this to OpenAI’s direct API (as of 2026): GPT-4o costs roughly $2.50/¥18 per million input tokens. TimiCC undercuts by 18x on input and 9x on output. | ||
| Claude’s direct API pricing sits around $3/¥21 per million input tokens. TimiCC’s ¥1 input rate is 21x cheaper. | ||
| The catch: these are reverse-proxy accounts, not official API keys. Latency averages 442ms — acceptable for most chat workloads, but slower than direct API calls (typically sub-200ms). |
Hidden Fees & Batch Discounts
No batch discount thresholds are published. No minimum top-up amount is specified. This is unusual — most relay stations require ¥10-¥50 minimum recharge. TimiCC lets you burn through your ¥1 credit and stop, no commitment. There’s no refund policy listed. If you recharge and the service degrades, you’re likely stuck. The “GPT Team accounts at 0.05x” pro means you can access GPT-4 Team tier models at 5% of their normal cost. For older GPT-4 models (not GPT-4o), this is essentially free.
Payment Methods
- Alipay (Alipay)
- WeChat Pay (WeChat Pay) Both are standard for Chinese users. No international credit cards accepted.
Pros & Cons
Pros:
- GPT-4o input at ¥1/million tokens — 18x cheaper than direct API
- Claude 3.5 Sonnet input at ¥1/million tokens — 21x cheaper
- ¥1 free credit with no credit card required
- 100% uptime recorded (small sample size caveat)
- Speed rating 4.5/5 — fast enough for production chat Cons:
- Only two models available (GPT-4o, Claude 3.5 Sonnet)
- QQ email registration only — privacy concern
- No refund policy specified
- No batch discount thresholds published
- Safety rating 3/5 — questionable data handling
- Composite score 60/100 — below average for relay stations
Verdict
TimiCC is for budget-first developers who need cheap GPT-4o and Claude 3.5 Sonnet access and are willing to accept registration quirks and limited model selection. The ¥1/million token input rate is genuinely competitive — you won’t find lower pricing from any legitimate relay station in 2026. But the QQ-only registration, missing refund policy, and low safety score mean this isn’t a platform for production workloads handling sensitive data. Use it for personal projects, prototyping, or cost-sensitive batch jobs where data privacy isn’t critical.
FAQ
Q: Does the ¥1 free credit expire? A: The platform doesn’t specify an expiration date. Assume it lasts indefinitely until used, but don’t rely on it for long-term storage. Q: Can I use TimiCC without a QQ email? A: No. Registration requires a QQ email address. This is a privacy concern — QQ accounts are tied to Tencent’s ecosystem and can be linked to your real identity. Q: What happens if I hit the 100,000 token limit? A: The platform lists a max token limit of 100,000 per request. If your prompt + completion exceeds this, the request will be rejected. Split large tasks into smaller chunks. Q: Is there a minimum recharge amount? A: Not specified. You can likely recharge any amount via Alipay or WeChat Pay, but without a published minimum, assume ¥1 or ¥10 as a practical floor. Q: How does TimiCC compare to OpenRouter for Claude 3.5 Sonnet? A: OpenRouter charges roughly $2/¥14 per million input tokens for Claude 3.5 Sonnet. TimiCC charges ¥1 — 14x cheaper. But OpenRouter offers 200+ models, refunds, and credit card support. TimiCC wins on price, loses on everything else.
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.
