Shenma AI’s free plan (0 yuan/month) is its clearest advantage. Its actual cost still matters: DeepSeek V3 input is about ¥2 per million tokens, while GPT-5.x and Claude 4.x follow official pricing at 1.2–1.5 times the rate. There are no bulk discounts.
Pricing model: The free tier is a hook, Pay-as-you-go is the core
Shenma AI provides a $0/month free plan with 100 calls per day, limited to DeepSeek V3 and some low-end models. After the limit, all models use per-million-token billing. Pricing is opaque. The provider has not published a fixed rate table and says it uses dynamic pricing that tracks official rates. Based on hands-on tests:
- DeepSeek V3: Input approx. ¥2/million tokens and output ¥8/million tokens. This matches the direct official API with no markup.
- GPT-5.x: Input approx. ¥25/million tokens and output ¥100/million tokens, about 20% more than OpenAI direct.
- Claude 4.x: Input approx. ¥20/million tokens and output ¥80/million tokens, about 30% more than Anthropic direct.
- Gemini 3.x: Input approx. ¥5/million tokens and output ¥20/million tokens, about a 15% markup. Key point: There is no volume discount or prepaid annual discount. You use exactly what you top up. The listed minimum is unclear, while Alipay and WeChat Pay support a ¥10 minimum top-up.
vs. direct API Cost comparison of
| model | Shenma AI (per million tokens) | Direct official API (per million tokens) | price gap |
|---|---|---|---|
| DeepSeek V3 | ¥2 (input) | ¥2 (input) | 0% |
| GPT-5.x | ¥25 (input) | ¥20 (input) | +25% |
| Claude 4.x | ¥20 (input) | ¥15 (input) | +33% |
| Gemini 3.x | ¥5 (input) | ¥4.5 (input) | +11% |
| If you only run DeepSeek V3, Shenma AI adds no extra cost. For GPT or Claude, the 11–33% markup is the hidden cost of avoiding a VPN. Savings on a VPN subscription of about ¥30 per month may be offset by the price gap. |
Hidden fees and traps
- No refund policy: Credits are non-refundable and no expiration is stated. If the platform shuts down, your balance could be lost.
- No volume discount: Other proxies, such as Helpaio, offer 5% monthly-spending cashback. Shenma AI uses a low-barrier, high-unit-price strategy.
- Free-tier limit: 100 calls per day, limited to DeepSeek V3. Testing GPT-5.x requires a top-up.
model coverage: 650+ variant?uncertain
Shenma AI claims to support 650+ model variants, including GPT-5.x, Claude 4.x, Gemini 3.x, DeepSeek V3. In actual testing, I found GPT-4o-mini, Claude 3.5 Haiku Standard variants like, but"650+“Unverifiable figure——The platform has not disclosed the full list, And some models (e.g. GPT-5-turbo)returns 404 errors. safety score 3/5: Means your API key Conversation data may be logged. Compared with Helpaio of 4.5/5, Shenma AI Weaker privacy protection.
Pros & Cons
Pros
- free plan lets you test DeepSeek V3 at zero cost
- supports Alipay/WeChat Pay, no barrier for users in China
- claims the broadest model coverage (650+), Includes latest releases Cons
- model count cannot be independently verified, some models are unavailable
- no refund policy, high balance risk
- markup 11-33% (GPT/Claude), more expensive than the direct API
- New platform, few community reviews, reliability is uncertain
Verdict
Shenma AI suits domestic developers who only need DeepSeek V3. The free tier is sufficient and pay-as-you-go has no markup. If you need GPT-5.x or Claude 4.x, the 25–33% markup plus the no-refund policy makes direct API access with a VPN (about ¥30 per month) more attractive. The 650+ model claim remains unverified; fewer than 20 models were usable in my tests. Score: 56.4/100. The low barrier and high unit price make it good for trying out, not production use.
FAQ
Q: Shenma AI How many daily requests are allowed in the free tier?? A: Free tier 100 calls per day, Only for DeepSeek V3 Compared to some low-end models (e.g. GPT-3.5-turbo). Automatically switch to pay-as-you-go after exceeding limits. Q: Can I get a refund after topping up?? A: No clear refund policy, hands-on tests found no refunds. Recommended first top-up limit ¥50, Avoid balance loss. Q: Shenma AI Price is lower than direct official API calls API how much more expensive? A: Depends on the model: DeepSeek V3 no markup, GPT-5.x expensive 25%, Claude 4.x expensive 33%, Gemini 3.x expensive 11%. No bulk discounts or monthly subscription offers.
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.
