Article

How to use Qwen4-Max’s free API quota

For developers trying Qwen4-Max, see how its free quota is restricted to Singapore, set a usage stop and understand fees outside the allowance.

Editorial illustration for How to use Qwen4-Max’s free API quota: a key and controlled doorway to a model. Not documentary evidence.

Qwen4-Max, released on November 16, 2026, offers eligible developers and small teams one million free tokens through QwenCloud’s Singapore pay-as-you-go API. Enable Free quota only before connecting an app, and budget separately for tool fees.

An API connects your software to the model; tokens are the units used to count its input and output. This is a temporary evaluation allowance, not unlimited free access. The November 11 Qwen4-Max preview was Token Plan-only, a separate access route.

1. Check your allowance and expiry

Sign up and complete payment verification. Select Singapore in the console, then open Free Tier → Model Invocation Quota. Find qwen4-max and check its balance and expiration date.

Input and output share one balance. The usual 90-day validity starts at account activation for existing models, or at release for models introduced after signup. Inactivity does not extend it.

Confirm the entitlement in your own account before building an integration. Selecting Singapore as a service region does not establish whether signup is available in your country.

2. Enable the stop before making calls

Turn on Free quota only in the model’s Actions column; it is off by default. QwenCloud says it blocks calls after exhaustion. Without it, tokens beyond the allowance can be charged within a request. If the switch is missing or disabled, check the model’s status first.

The quota display can lag several minutes. Do not treat it as a live spending meter.

This control concerns model usage, not the entire application bill. Built-in web and image search fees are outside the token allowance. Leave paid tools disabled for a token-only evaluation; external services and hosting need their own budgets.

3. Match the key, endpoint and model

With Singapore selected, open API Keys and choose your workspace. In the pay-as-you-go section, select Create API key, add a description and choose Generate Key. The full key is displayed only once; save it in your application’s protected server-side secret store, not browser code.

Use the Singapore configuration:

  • Region: ap-southeast-1.

  • Chat Completions base URL: https://ap-southeast-1.maas.qwencloudapi.com/compatible-mode/v1.

  • Released model ID: qwen4-max, not qwen4-max-preview.

The key and endpoint must belong to the same region. Changing the console selector does not rewrite an existing app’s configuration. The regional documentation also distinguishes data storage from processing: Singapore stores data there, but its International inference scope uses nodes outside mainland China, not necessarily only Singapore.

Do not substitute a Token Plan key. Token Plan consumes separate Credits, not this trial balance, and its exhausted allowance does not automatically become pay-as-you-go usage.

4. Start with a small text-only request

Qwen4-Max has thinking enabled by default, and those reasoning tokens count as billed output. A simple connection check can disable thinking and search before you evaluate more demanding tasks.

Illustrative HTTP JSON body, derived from the Chat Completions reference and not executed here. Send it to /chat/completions under the base URL above, with authentication supplied by your server-side secret mechanism:

{
  "model": "qwen4-max",
  "messages": [
    {
      "role": "user",
      "content": "Suggest three headings for a project plan."
    }
  ],
  "max_completion_tokens": 512,
  "enable_thinking": false,
  "enable_search": false
}

The 512-token setting is an illustrative choice for a short reply. The output limit covers reasoning plus the answer; the reference allows variation of up to 10 tokens. These are direct HTTP fields: in the compatible Python SDK, the thinking and search options instead belong in extra_body.

Inspect usage.prompt_tokens, usage.completion_tokens and usage.total_tokens in the response. Reasoning is already part of output usage—do not add its breakdown a second time. Visible answer length alone is not a cost measure.

If you receive 403-AllocationQuota.FreeTierOnly, the error reference identifies an exhausted allowance with free-only mode as one possible cause. Handle that as a stop, not a reason to automatically disable the control or switch to a paid fallback.

5. Calculate the bill before allowing paid usage

As of November 16, Singapore’s Qwen4-Max list rates are US$1.50 per million input tokens and US$4.50 per million output tokens. Thinking and non-thinking use the same unit prices, with no input-length tiers.

For an illustrative workload after the free allowance no longer applies:

  • 100,000 uncached input tokens: 100,000 ÷ 1,000,000 × $1.50 = $0.15.

  • 20,000 billed output tokens, including any thinking: 20,000 ÷ 1,000,000 × $4.50 = $0.09.

  • Total model cost: $0.24.

This calculation assumes those are the complete token totals. It excludes cache operations, tool calls, retries, taxes and hosting; account-specific discounts or currency conversion can change the final bill. Do not apply a generic batch discount: Qwen4-Max does not support batch inference.

Tools can cost more than that small model workload. Singapore’s built-in web search costs $10 per 1,000 calls, so 100 billable searches would cost $1 separately—even while model tokens remain free. That is a calculation for 100 searches, not 100 conversations.

If you add a separate retrieval service, assess its allowance too. Our Firecrawl MCP free-access and usage-limits guide covers account requirements and tool limits.

After the first requests, inspect Analytics → Usage and the pay-as-you-go bill breakdown by model, key and workspace. Usage follows the selected region; billing covers all regions and can lag. An empty immediate bill is not proof that no charge is accruing.

Keep the trial’s stop in place until you deliberately choose paid continuity and a budget. If you are comparing access routes, the Gemini 4 Hydrogen API access and pricing guide explains another model’s trial and paid options.