If you want to run your own complex tasks on Qwen4-Max, there is now an application route for preview access. It is not an unrestricted chat launch. Qianwen AI Platform opened its evaluation on November 11, with a November 18 deadline and 100 places reviewed in submission order. Applications may close early once the places are filled. Reports and supporting materials must also be submitted before the activity ends on November 18; applying by then is not enough. Prepare a task before applying for API access.
Choose the reporting workload, not just the allowance
The evaluation page offers three tracks. The entry track provides one hundred million tokens for at least one complex case from a real work setting. The deeper track provides two hundred million tokens and requires execution data for each case. The professional track provides five hundred million tokens and requires a self-built evaluation set, a complete report and execution files.
The corresponding vouchers are worth 100, 300 and 500 yuan. These are not cash payments for signing up. An approved application grants the testing allowance; a successful review of the submitted materials earns the voucher. Choose the track whose documentation you can actually finish, rather than the one with the largest allocation.
An API workflow with explicit limits
Qwen4-Max-Preview accepts image and video inputs and offers a thinking-mode switch. The activity lists a 1M context window, maximum concurrency of five and a default rate limit of 300W TPM. W denotes ten thousand, and TPM measures tokens per minute, not requests per minute. These specifications do not establish that the model will accurately use every detail of a long input. The allowance is not a service commitment for high-concurrency workloads.
To start, select a track, read the agreement and apply. After approval, obtain an API key and configure it through an environment variable. The page supplies Python, Node.js and cURL examples using an OpenAI-compatible interface. Its sample sets base_url to https://qweneval.aliyuncs.com/v1 and model to qwen-latest-series-invite-1102. Copy the code from the original page; do not substitute the display name for the request parameter or reuse an old project's endpoint without checking it.
Budget time for the evidence, not just the test
Both testing and report preparation need to fit inside the evaluation window. Application approval grants access, but participants must still submit their results, reports and supporting materials for the chosen track before the activity ends on November 18. The deeper and professional tracks require execution files, so keeping only the final answers will not satisfy those requirements.
The organizer warns that the preview will keep changing. Our recommendation remains to save inputs, outputs, execution records and test dates as you go. Validate your own tasks before considering a migration, and do not treat a free testing allowance as a commercial price quote.