Procurement / Sep 6, 2026

Price-Only Chinese AI API Procurement Fails

Use a source-dated procurement checklist for Chinese AI API routes instead of ranking vendors by a single token price.

Keyword report: 2026-09-05Tier 1/2 developer focusSources checked Sep 6, 2026

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 includes direct model and pricing interest, but the most useful article for Tier 1 and Tier 2 buyers is not another price ranking. Chinese AI API procurement fails when a team sorts by one input-token row and ignores output length, cache behavior, tool fees, failed-call policy, account group, data boundary, support path, and route availability.

This guide is a risk checklist for finance, security, and engineering reviewers. It uses live AIWave rows and current provider documentation as source-dated evidence, then turns those facts into a procurement packet. The outcome is a repeatable review that can approve a small trial, reject a poor-fit workload, or request more evidence without pretending that a single price number answers the whole buying question.

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 listed AIWave gateway rows such as DeepSeek V4 Flash at $0.638 input, $0.0202884 cache-hit input, and $1.914 output per 1M tokens, plus DeepSeek V4 Pro at $1.914 input, $0.0637362 cache-hit input, and $5.742 output.

DeepSeek official pricing checked on Sep 6, 2026 lists per-1M-token V4 rows split by cache-hit input, cache-miss input, output, and UTC peak/off-peak windows. Flash rows include $0.007 off-peak and $0.014 peak cache-hit input, $0.22 off-peak and $0.44 peak cache-miss input, and $0.66 off-peak and $1.32 peak output. Pro rows include $0.022/$0.044 cache-hit input, $0.66/$1.32 cache-miss input, and $1.98/$3.96 output.

Z.AI, QwenCloud, and Kimi pages checked on Sep 6, 2026 show why token-only procurement is incomplete. Z.AI lists GLM model rows plus tool and media costs, with GLM-5.3-Flash promotional rows ending on September 9, 2026 Singapore time. QwenCloud documents Batch API at 50% of real-time token pricing, context caching, thinking tokens billed as output, and failed-call behavior. Kimi lists web search at $0.004 per invocation and context caching for repeated content.

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.

Procurement fieldBad shortcutSafer evidence
Input priceSort by one input rowStore input, cache-hit, cache-miss, and output rows
OutputAssume short answersSet task-specific output caps
CacheAssume all repeated context benefitsMeasure cache share by prompt template
ToolsHide search and media inside token costSeparate per-use and per-asset fees
Failed callsIgnore failed-call policyRecord provider policy and retry class
Gateway groupForget account multiplierStore default or VIP group with request evidence
SecuritySend production data too earlyRun a redacted canary first

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 dataclasses import dataclass

@dataclass
class ProcurementRow:
    row_owner: str
    model: str
    source_url: str
    checked_at: str
    input_per_1m: float
    output_per_1m: float
    cache_hit_per_1m: float | None
    account_group: str

deepseek_flash_aiwave = ProcurementRow(
    row_owner="aiwave_gateway",
    model="deepseek-v4-flash",
    source_url="https://aiwave.live/pricing",
    checked_at="2026-09-06",
    input_per_1m=0.638,
    output_per_1m=1.914,
    cache_hit_per_1m=0.0202884,
    account_group="VIP",
)

print({"api_key": "YOUR_API_KEY_HERE", "row": deepseek_flash_aiwave})

Start With Row Ownership

Every price row should identify the owner of the row. A direct-provider row belongs to DeepSeek, Z.AI, QwenCloud, or Kimi. An AIWave row belongs to the gateway pricing surface and current account group. A cloud marketplace row has its own terms. If the worksheet cannot tell row ownership from a glance, the procurement packet is not ready for approval.

Store Token Classes Separately

Input, cache-hit input, cache-miss input, and output should remain separate columns. A single blended number can be useful after the workload is measured, but it should not be the source of truth. Long-context agents, RAG systems, and coding reviews often have large input and variable output. The same average price can hide very different operational risks.

Add Tool and Media Columns

Model families increasingly include search, image, video, audio, OCR, and agent features. Z.AI and Kimi publish per-use or per-asset rows for some of those features, and QwenCloud documents several billing modes. Put tool and media costs beside token costs, not inside prose. A procurement reviewer should be able to disable search or media features without changing the base chat route.

Record Failed-Call Policy

Failed-call handling affects both budget and reliability. QwenCloud explicitly documents failed-call behavior, while other routes may expose different billing and retry semantics. The trial receipt should classify authentication failures, provider rejection, capacity response, timeout, context-length error, and user cancellation. Retrying every failure class in the same way can inflate cost and hide the real fix.

Tie Price to Data Boundary

The risk checklist should include the data classes allowed in the trial. A redacted canary can prove API shape, model availability, usage fields, and support path before sensitive production data is sent. Security should approve the data boundary separately from finance's approval of a price row. A route can have an attractive rate and still be a poor fit for regulated or contract-bound workloads.

Procurement Review

A complete packet includes source URL, checked date, row owner, model ID, token classes, account group, output cap, tool policy, failed-call policy, data boundary, support path, and recheck owner. Engineering should attach a redacted receipt. Finance should preserve direct-provider and gateway rows separately. Security should document what is excluded from the first trial.

Final Checklist

A price-only ranking should become a route-risk worksheet before any production decision. Keep current prices source-dated, preserve token classes, measure cache behavior, cap output, govern tool calls, classify failed requests, and keep data exposure small until the route is approved. Recheck all provider pages and AIWave /api/pricing before a purchase order or recurring budget forecast.

Source Links

Related AIWave Links