AI Use Cases: What to Build with Chinese AI Models

These are the workloads developers actually ship on the AIWave API: chat assistants, coding tools and retrieval-augmented search over private documents. Each guide covers the architecture, the model choice, working code and what the workload costs at real volumes — all through one OpenAI-compatible endpoint covering DeepSeek, GLM, Kimi, Qwen and ERNIE.

Guides

Build a Customer Service Chatbot with Chinese AI Models

Build a multilingual customer service chatbot using DeepSeek and GLM. Step-by-step guide with code examples. 10× cheaper than GPT-4.

Build an AI Coding Assistant with DeepSeek

Wire DeepSeek V4 Pro into your editor for completion and review. Setup steps, prompt patterns, and what it costs per developer each month.

Low-Cost RAG with Chinese Embedding Models

Build a Retrieval-Augmented Generation (RAG) pipeline using Chinese AI embedding and chat models. Complete guide: chunking, vector DB, retrieval.

Choosing a model for the job

Use-case fit is mostly a price-versus-capability trade-off, and because model IDs are just strings here, you can change your mind later without rewriting anything.

Current rates for the models these use cases most often reach for:

Model ID	Input / 1M tokens	Output / 1M tokens
deepseek-v4-flash	$0.206	$0.412
deepseek-v4-pro	$1.09	$2.17
glm-4.7-flash	free	free
glm-5	$1.55	$4.96
kimi-k2.5	$0.66	$3.30
qwen3-coder-480b-a35b-instruct	$0.12	$0.36

Rates read from the AIWave pricing endpoint on 2026-07-26. Check live pricing before budgeting — providers revise rates.

• Chat assistants are latency-sensitive and high-volume — a cheap fast model usually wins on user experience as well as cost.
• Coding tools reward capability; a code-specialised model earns its price on diff quality rather than raw token cost.
• RAG is dominated by input tokens, since every query drags retrieved context along. Long-context models reduce chunking complexity but cost more per call.

Patterns that apply to all three

• Route by difficulty so premium models only see hard requests.
• Stream responses — perceived latency matters more than total latency in interactive products.
• Handle 429 and 503 properly and keep a fallback model configured.
• Use function calling when the model needs to reach real systems rather than guess.

Where to go next

• API documentation — base URL, auth, endpoints, error codes
• Model directory · Live pricing
• Access Chinese AI models without a Chinese phone number
• Provider comparisons

Sources

Every claim above traces back to one of these. Pricing figures come from the AIWave pricing endpoint on 2026-07-26; capability claims come from the vendors’ own documentation.

• OpenAI Chat Completions API reference
• AIWave API documentation