This guide uses source checks from Sep 6, 2026. Provider and gateway prices can change; preserve the checked date with every forecast.
Why This Topic Matters Now
The Sep 5 keyword report is still dominated by brand and site queries from Tier 1 markets. `aiwave.live` produced 57 impressions with 56 from Tier 1 countries, and `site:aiwave.live` produced 49 Tier 1 impressions with no clicks. That pattern points to evaluation behavior: people are checking whether AIWave is a real technical route, not asking for another broad model ranking.
For a Tier 1 engineering team, the relevant comparison is not only a model price row. It is the operational work behind nine provider accounts: identity, billing, currency, API clients, model names, usage receipts, route failures, security review, and support. This guide turns that evaluation into a worksheet for teams comparing direct Chinese AI provider accounts with one OpenAI-compatible AIWave gateway.
Source Facts Checked Today
AIWave /api/pricing checked on Sep 6, 2026 returned success=true, 63 records, pricing_version a42d372ccf0b5dd13ecf71203521f9d2, default group ratio 3, and VIP group ratio 1. The public pricing page checked the same day showed the current catalog rows across DeepSeek, GLM, Kimi, Qwen, ERNIE, MiniMax, Doubao, StepFun, and MiMo, with DeepSeek reference rows dated 2026-08-27.
The AIWave pricing page checked on Sep 6, 2026 listed DeepSeek V4 Flash at $0.638 input, $0.0202884 cache-hit input, and $1.914 output per 1M tokens. It listed DeepSeek V4 Pro at $1.914 input, $0.0637362 cache-hit input, and $5.742 output. The same page showed examples such as GLM-5.1 at $2.10 input, $0.680001 cache-hit input, and $6.60 output, Kimi K3 at $4.50 input, $0.90 cache-hit input, and $22.50 output, and qwen-72b-chat at about $4.46342 input and output.
Official provider pages checked on Sep 6, 2026 expose different operational assumptions. DeepSeek publishes OpenAI and Anthropic base URL formats with V4 peak/off-peak token rows and route concurrency limits. Z.AI publishes GLM text, tool, media, and agent rows. QwenCloud documents Batch API, context caching, thinking-token billing, and failed-call behavior. Kimi documents token billing, a $0.004 web-search add-on, and context caching.
Planning Matrix
A source-dated planning matrix keeps the page useful for engineers and procurement reviewers. It turns a search query into an auditable route decision instead of a loose model preference.
| Operational area | Nine direct accounts | One gateway route |
|---|---|---|
| Identity | Multiple workspaces, owners, and recovery paths | One AIWave account owner plus internal approvers |
| Billing | Different currencies, invoices, and price tables | One USD ledger with source-dated model rows |
| API client | Provider-specific SDK or base URL variants | OpenAI-compatible base URL and model switch |
| Model naming | Aliases differ across providers and clouds | Route names listed in one catalog |
| Usage evidence | Usage fields and receipts vary | One expected request and ledger shape |
| Failure handling | Separate status pages and support paths | Gateway route policy plus provider caveats |
| Procurement | Nine vendor reviews if fully direct | Gateway review plus provider-risk notes |
Implementation Pattern
The implementation pattern keeps credentials as placeholders, pins the AIWave base URL, records the model, and leaves room for route-specific controls. Production applications should move credentials into environment or secret storage.
from openai import OpenAI
client = OpenAI(
api_key="YOUR_API_KEY_HERE",
base_url="https://aiwave.live/v1",
)
provider_routes = ["deepseek-v4-flash", "glm-5", "kimi-k3", "qwen-72b-chat"]
for model in provider_routes:
response = client.chat.completions.create(
model=model,
messages=[{"role": "user", "content": "Return one sentence for a redacted route check."}],
max_tokens=120,
temperature=0.1,
)
print({"model": model, "usage": response.usage})
Count Operational Surfaces
Provider breadth is only useful when the operational surface is visible. A direct setup with nine Chinese AI providers can mean nine account owners, nine billing dashboards, several currencies, multiple API shapes, different support paths, and separate model-name conventions. A gateway setup concentrates that work, but it does not erase provider differences. The worksheet should count which surfaces move to AIWave and which risks still belong to the underlying route.
Keep Row Ownership Visible
The pricing workbook should label every row as direct-provider, cloud marketplace, or AIWave gateway. Direct DeepSeek, Z.AI, QwenCloud, and Kimi pages are useful reference sources, but they are not automatically the same as a gateway row. AIWave rows checked on Sep 6, 2026 come from the AIWave pricing API and pricing page. Provider rows come from provider documentation. Mixing them into one blended number makes procurement harder to defend.
Measure Integration Work
A team replacing provider-specific SDKs should measure the work in base URLs, auth headers, model names, streaming behavior, error shape, retry policy, usage object, and ledger export. If the OpenAI client remains stable and only the model string changes, that is a real integration advantage. If each provider still needs special code paths, the gateway should document those exceptions before production traffic begins.
Define a Trial Boundary
The first trial should avoid sensitive production data. Use a redacted prompt, a synthetic ticket, or a public task. Record route owner, model ID, input tokens, cached input tokens when visible, output tokens, response status, retry class, output cap, pricing source, and checked date. A successful answer without a receipt is a demo; a successful answer with a source-dated receipt is procurement evidence.
Plan Fallback by Behavior
Fallback should not silently change model behavior. A DeepSeek planning route, a GLM reasoning route, a Kimi long-context coding route, and a Qwen batch route can all respond through a compatible client shape, but they may differ in context window, output style, tool behavior, and billing fields. Store fallback eligibility by task class rather than letting a generic retry handler pick any available model.
Internal Links for Evaluators
Readers should move through Chat Completions, Models docs, Pricing, Trust, Privacy, and the quickstart evidence pack. The goal is to make the brand query lead to a working evidence route, not a disconnected page visit.
Procurement Review
Procurement should ask for the provider account count avoided, the AIWave account owner, the current pricing_version, the group ratio used in examples, the model routes tested, the source dates, and the support path. Engineering should attach redacted receipts. Security should attach the data boundary. Finance should preserve direct-provider rows separately from AIWave rows so later price movement can be reviewed without reconstructing the trial.
Final Checklist
Choose the gateway route only when it reduces real operational work: fewer accounts, stable OpenAI-compatible client code, clearer model naming, a single ledger, dated price evidence, and controlled fallback. Keep direct-provider rows as reference evidence. Recheck AIWave /api/pricing and provider pages before purchase approval, because the point of the worksheet is current, auditable evidence rather than a permanent claim.