GPT-4o input costs ¥0.5 per million tokens on Terminal.Pub, and output costs ¥2.5 per million — roughly 1/3 of OpenAI’s direct pricing for the same model. This isn’t a production relay station. It’s a budget testing sandbox for developers who want to poke around GPT-4o and Claude 3.5 Sonnet without committing to a monthly subscription. The platform launched in February 2026, so it’s fresh — , unproven.
Pricing Breakdown
Terminal.Pub only offers two models: GPT-4o and Claude 3.5 Sonnet. No DeepSeek, no Gemini, no Llama. That’s it.
| Model | Input (per million tokens) | Output (per million tokens) |
|---|---|---|
| GPT-4o | ¥0.5 | ¥2.5 |
| Claude 3.5 Sonnet | ¥0.5 | ¥2.5 |
| Compare that to OpenAI’s direct pricing: GPT-4o costs about ¥1.5 input / ¥6 output per million tokens. Terminal.Pub undercuts by 66% on input and 58% on output. | ||
| Bottom line: You’re paying roughly 1/3 of direct API rates, but you’re getting 87.5% uptime and 1134ms latency — not production numbers. |
Free Trial & Payment Details
Terminal.Pub offers a free tier (no credit card required — just register). You get free API groups to test both models. No promo code exists. Payment methods: Alipay and WeChat Pay only. No credit cards, no PayPal, no crypto. That’s fine for Chinese developers, but a blocker for anyone outside the WeChat/Alipay ecosystem. Hidden fees: The platform doesn’t specify a minimum recharge amount. There are no batch discount thresholds either — the ¥0.5/¥2.5 rate applies at any volume. No per-request overhead fees are listed. Key takeaways:
- Free tier requires no credit card — just sign up
- Only Alipay and WeChat Pay accepted
- No minimum top-up amount disclosed
Pros & Cons
Pros:
- Extremely low pricing — GPT-4o at ¥0.5 input / ¥2.5 output
- Free API groups for testing both models
- Supports Claude and GPT without VPN Cons:
- Very new (Feb 2026) — unproven long-term stability
- 87.5% uptime and 1134ms latency — not production-grade
- No monthly subscription option
- Only two models available
- Such low pricing raises sustainability concerns
- No refund policy specified
Performance Reality Check
Uptime: 87.5%. That’s roughly 3 hours of downtime per day. For a testing sandbox, that’s tolerable. For anything customer-facing, it’s a non-starter. Latency: 1134ms average. That’s over a full second per request. Direct GPT-4o API calls typically return in 300-600ms. The extra half-second comes from relay routing. Safety rating: 3/5. Speed rating: 3.5/5. Composite score: 52.5/100. These numbers tell the story: this is a budget option, not a premium service. Key takeaways:
- 87.5% uptime means ~3 hours daily downtime
- 1134ms latency is 2-3x slower than direct API
- Composite score of 52.5/100 reflects the trade-off for low price
Verdict
Terminal.Pub is for one thing only: cheap, quick testing of GPT-4o and Claude 3.5 Sonnet without a VPN. If you’re a Chinese developer who wants to verify a prompt works before deploying to production, this works. But don’t build your app on it. The 87.5% uptime and 1134ms latency disqualify it for production use. The lack of model variety means you can’t experiment with Gemini or DeepSeek. And the platform’s age (Feb 2026) means it hasn’t survived a real outage or abuse attack yet. Would I use it? For a weekend project or a one-off benchmark, sure. For anything that pays my rent, no.
FAQ
Does the free trial require a credit card?
No. Terminal.Pub’s free tier activates on registration with no payment method required. You get free API groups immediately.
What happens if I exceed the free tier limits?
The platform doesn’t specify a hard cap or throttling policy. Given the low pricing, expect either rate limiting or a request to recharge via Alipay or WeChat Pay.
Can I use Terminal.Pub for production workloads?
Not recommended. 87.5% uptime means ~3 hours of downtime daily. 1134ms latency is 2-3x slower than direct API calls. This is a testing sandbox.
Are there batch discounts for high-volume usage?
No. The ¥0.5 input / ¥2.5 output per million tokens rate applies at any volume. No batch discount thresholds exist.
How does Terminal.Pub compare to direct OpenAI API pricing?
Terminal.Pub charges roughly 1/3 of OpenAI’s direct rates for GPT-4o: ¥0.5 vs ¥1.5 input, ¥2.5 vs ¥6 output. You trade price for reliability and latency.
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.
