Baidu Qianfan API Pricing: International USD Listings Versus Mainland RMB Tariffs

Start with the correct contracting channel

Baidu AI Cloud’s international Qianfan campaign page displays ERNIE 5.0 at USD 1.40 per million input tokens and USD 5.60 per million output tokens. Its model list, updated June 25, 2026, identifies the ernie-5.0 endpoint and separately lists model limits. Those documents provide evidence for the international Qianfan channel, not blanket permission to purchase through a mainland China account at those USD prices. Mainland price tables can instead use yuan per thousand tokens, so multiplying by 1,000 correctly is essential before comparing numeric rates.

Keep provider and marketplace rates distinct

The international page also advertises selected third-party models, including DeepSeek and GLM. A DeepSeek tariff on Baidu infrastructure is a Qianfan channel price; it is not necessarily the same as a contract signed directly with DeepSeek. The headline rate may depend on the precise API endpoint, model revision and promotional campaign. A global reference table should identify provider, reseller, region, currency, model ID, cached/uncached input, output, and whether its source is an enduring catalog or a special-offer landing page.

Worked token calculation with an explicit scope

At the international ERNIE 5.0 displayed list rates, 800,000 uncached input tokens and 200,000 output tokens would calculate to USD 2.24 before taxes and other metered services: 0.8 × 1.40 plus 0.2 × 5.60. This is an arithmetic demonstration, not an observed bill or a claim that every customer receives access. If a task adds retrieval, caching or storage, those components must be priced from their own documented meters. A cost tool should refuse a cross-currency comparison without a dated foreign-exchange quote.

Rate limits belong beside the price

The vendor’s model list states default RPM and TPM limits for specified endpoints. Request throughput and token throughput can constrain a production rollout long before monthly budget is exhausted; read the exact row instead of assigning one universal Qianfan ceiling. A model with a lower price per million tokens may be operationally unsuitable if its available tokens per minute, supported input format, or activation requirements prevent a service-level objective. Log observed 429s separately from price so a retry burst is not mislabeled a tariff increase.

What should remain unverified

The original start date of the visible international USD price is not specified in the linked live card. Save its checked-at timestamp but leave effective_at empty. Do not quote current China-region checkout amounts, taxes, promotional expiration dates or service eligibility unless an exact first-party document confirms them for the account. Compare alternatives with equal token distributions and channel constraints rather than converting a marketing price into a universal procurement promise.

Decision checklist and limitations

Before purchase or deployment, retain the literal model or plan identifier, payer/merchant, country or cloud region, billing currency, exact usage meter, source URL, checked date, effective date if explicitly supported, and any plan commitment or promotional eligibility. Recalculate from the companion price and quota snapshots rather than assuming that a remembered list tariff will remain unchanged. A credit allowance, maximum input length, published token rate and confirmed billed currency answer different questions; never add them as if they were the same quantity.

When comparing another provider or route, use the same input/output distribution and acceptable result definition. Include unsuccessful attempts, optional tools, storage, tax, payment and human review only when those components are actually measured or supported by an explicit source. Vendor marketing and documentation are not independent product tests. This article makes no claim to have personally tested paid endpoints, purchased plans, or checked account-specific taxes or availability.