Shenma Relay API Review: Pricing & Comparison

What does the number 2557 actually mean

Competition on model count among AI API relays is fierce — but the “2557” figure needs some context. Industry convention counts different version snapshots of the same model separately (e.g. gpt-4o-2024-11-20 and gpt-4o-2025-01-29 count as two), so the actual set of distinct model capabilities is usually well below the advertised number. Even so, Shenma Relay API’s coverage still exceeds most platforms, and the key differentiator is breadth across modalities:

Modality	Representative models
Text conversation	Claude Fable 5, GPT-5.5, Gemini 3.1 Pro, DeepSeek V4, Grok 4.3
Image generation	Midjourney v7, Nano Banana 2, DALL-E 3
Video generation	Sora 2, Kling
Audio/voice	The Whisper series

This “text + image + video + audio all in one place” coverage has real value for dev teams building multimodal applications at the same time — no need to maintain four separate accounts and API keys.

Why someone might pick Shenma over a specialized platform

Shenma Relay API’s positioning isn’t “most stable” or “cheapest” — it’s breadth of coverage. Specific scenarios where that matters:

Needing the Midjourney API: Midjourney still has no officially open public API, so going through an aggregation platform is one of the mainstream workarounds. Shenma is one of the platforms offering this kind of endpoint.

Needing the Grok lineup: xAI’s Grok 4.3 has limited mainland direct-connect channels, and Shenma is one of the platforms with publicly documented access to it.

Validating multimodal workflows: at the product-prototyping stage, when you need to quickly validate a full “text generation → image rendering → video synthesis” pipeline, running it through one account is more efficient than juggling multiple accounts.

Testing niche/newer models: models like the latest Llama 4 versions or Nano Banana 2 that are less common — specialized platforms often don’t carry them, and Shenma’s broad-coverage strategy makes it more likely to have them listed.

How to access the models

All models are accessed through an OpenAI-compatible interface — just swap the base_url:

from openai import OpenAI

client = OpenAI(
    api_key="your key",
    base_url="https://api.whatai.cc/v1"
)

# Text model
response = client.chat.completions.create(
    model="claude-fable-5",
    messages=[{"role": "user", "content": "Hello"}]
)

# Image generation (defer to the official docs for the exact Midjourney format)
response = client.images.generate(
    model="midjourney",
    prompt="a cat in space, photorealistic"
)

Note: request formats for image/video/audio generation differ from the text API — for multimodal endpoints, follow the specific conventions in the official docs (api.whatai.cc).

Pricing

Pay-as-you-go, billed in RMB, mainland direct connect. The top-up exchange rate is roughly ¥2 = $1 of credit (about 28% of official pricing); you get $0.2 in free credit on signup, and daily check-ins on the dashboard add more. Defer to the official site’s live pricing page for exact figures — given the large model count, billing methods differ by modality too (text by token, image by piece/resolution, video by second).

Make sure to check the specific unit price for your target model before topping up — billing variance on multimodal platforms is much larger than on pure-text relays.

An honest assessment

Shenma Relay API’s core value is model breadth; its weakness is that it doesn’t have the same level of stability track record as better-known names. If your primary need is high-quality, day-to-day Claude Code usage, a stability-focused option like KoalaAPI or 4SAPI would be a better fit; if you need multimodal coverage or want to test niche models, Shenma is worth trying.

Top up a small amount first to test your target model’s actual response quality and billing accuracy before committing to heavier use.

Compared with DMXAPI: DMXAPI’s 480+ models likewise span text + image + music + video multimodal, but it’s positioned more toward enterprise-grade stability, with more independent review data than Shenma. If your primary need is the text API and documentation completeness matters most, No.1-API’s API docs and code samples are more thorough.

Information verified 2026-08-11. The 2557 model count is a vendor claim; the multi-version-snapshot counting method may differ from the actual set of distinct model capabilities — defer to api.whatai.cc’s live data.

Related reviews

• Zhipu AI: Zhipu’s official GLM series, multimodal AI, top-tier Chinese-language capability
• ofox.ai: a unified LLM gateway, 100+ models including GPT/Claude/DeepSeek, focused on Claude compatibility and official direct-connect with no capability downgrade
• MegaLLM: 70+ models via official channels, stable and highly available, a global AI API relay for mid-to-high-end needs
• Replicate: open-source model aggregation plus custom model deployment, a leading platform for image/video generation, supports publishing private models