This guide uses source checks from Sep 23, 2026. Provider and gateway prices can change; preserve the checked date with every forecast.
Why This Topic Matters Now
The Sep 22 keyword report surfaced both `chat completion stream` and `responses api streaming`. They look like syntax queries, but the production problem is a contract boundary: a client can receive some tokens and still lose the usage record, terminal event, or retry context. A stream is not validated merely because text appears on screen.
This guide gives Tier 1 and Tier 2 teams a small streaming acceptance set for OpenAI-compatible Chinese AI routes. It keeps Chat Completions and Responses API semantics distinct, treats server-sent events as a transport contract, and records the live AIWave route check separately from the dated USD pricing snapshot. The objective is a client that can explain a completed stream, an interrupted stream, and a stopped stream without guessing.
Source Facts Checked Today
AIWave /api/pricing was checked from production on Sep 23, 2026 and returned HTTP 200, success=true, 73 live route rows, pricing_version a42d372ccf0b5dd13ecf71203521f9d2, auto_groups=['default'], group_ratio default=1 and vip=0.9. The public /api/v1/pricing endpoint returned HTTP 200 with 56 dated USD rows, pricing_version 83f77abde81ee3a096a672ed959ccc096f5d37a45c177ae8e03229456b5415a5, checked=2026-09-10, and updated_at=2026-09-18. Use the live response for route availability and the dated JSON for the USD forecast; they are not one interchangeable rate table.
The AIWave Chat Completions documentation and live pricing endpoint were both checked on Sep 23, 2026 and returned HTTP 200. The public docs provide a Chat Completions request path; this article does not infer that every Responses API event or field is supported through that path.
The dated AIWave pricing JSON checked in this run lists `deepseek-flash` at $0.70 input, $0.0233 cache-hit input, and $2.10 output per 1M tokens, effective 2026-09-10. Keep those dated USD values separate from the live route table and from any provider-direct peak or off-peak table.
MDN describes server-sent events as a one-way server-to-client stream using the EventSource model. For an SDK request, the practical checks are event framing, reconnection policy, terminal handling, and application-level idempotency; transport reconnection alone is not proof that the original model request is safe to repeat.
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.
| Contract field | Failure mode | Acceptance evidence |
|---|---|---|
| Content type | Client parses a buffered body | HTTP headers and first event |
| Delta ordering | Text is duplicated or reordered | Sequence counter and final text |
| Terminal event | UI closes without usage or finish reason | Explicit end-state assertion |
| Disconnect | Retry repeats a side effect | Request ID and idempotency policy |
| Usage | Budget misses streamed output | Final usage or documented absence |
| Timeout | Worker loops after partial output | Attempt ceiling and stop reason |
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",
)
stream = client.chat.completions.create(
model="deepseek-flash",
messages=[{"role": "user", "content": "Return one bounded line."}],
max_tokens=80,
stream=True,
)
parts = []
for chunk in stream:
delta = chunk.choices[0].delta.content or ""
parts.append(delta)
print({"delta": delta, "request_id": getattr(chunk, "id", None)})
print({"text": "".join(parts), "attempts": 1})
Turn the Query Into a Contract
For streaming API contract tests, write down the request shape, model ID, data class, output ceiling, timeout, retry ceiling, owner, and source date before the first trial. A compact contract gives engineering and procurement the same object to review when a provider changes a route, a model family, or a billing field.
Separate Live Routes From Dated Rates
The live AIWave pricing response tells you which route rows and endpoint types are available at the check time. The public pricing JSON is a dated USD snapshot for forecasting. Preserve both URLs, versions, checked dates, model IDs, and account-group context in the decision record instead of presenting a volatile source as a permanent quote.
Use a Small Acceptance Set
A useful canary covers a normal request, an empty or malformed request, a repeated prefix, a long output, a disconnect, and a deliberate stop condition. Record request ID, model ID, status, token usage, finish reason, retry count, and reviewer outcome. This turns a search result into evidence that can survive a route update.
Keep Data and Credentials Bounded
OpenAI-compatible clients reduce integration work, but they do not choose the right data boundary. Keep the credential server-side, use a visible placeholder in examples, redact fixtures, and attach a data-class decision to every route policy. Do not let a model alias, feature flag, or context mode silently widen what crosses the API.
Make Recovery Observable
Retry only failures that are safe to retry and cap every fallback. Preserve the original request ID, mark the stop reason, and distinguish provider errors from client validation, policy rejection, and budget stops. Silent loops hide both reliability failures and billing variance.
Use AIWave's Evidence Layer
Use the Chat Completions docs, Models docs, dated Pricing JSON, and Status. Recheck the live route table before rollout, the dated pricing JSON before a budget review, the status page before a launch window, and the trust page before procurement. Keep those checked dates visible in the internal record.
Release Gate
Promotion is ready when the provider source is dated, the AIWave route is rechecked, the acceptance set passes, the billing fields are understood, and a named owner can stop or reverse the change. If a field is unknown, label the work as a trial rather than production.