Qwen3.8 Flash API pricing and cost calculator

How this is calculated

The calculator starts with 30,000 input tokens and 3,000 output tokens per call. With half the input read from cache, one call costs $0.0039: $0.00249 for input and $0.00141 for output. With no cache hits it costs $0.00591, and with all the input read from cache it costs $0.00189. At 10,000 calls a month the chart's starting total is $39. Now try a chat product. Suppose each turn sends 6,000 input tokens, most of them a system prompt and history that come from cache, and returns 700 output tokens. At 500,000 turns a month that costs $292.90. A document or video pass looks different. A job with 800,000 input tokens and a 4,000 token summary costs $0.12188 with no cache hits. How many tokens a video or an image takes depends on its length and resolution, so test your own files before you multiply anything by a thousand. Reasoning is the part that surprises people. A hard problem with 10,000 input tokens that produces 20,000 output tokens costs $0.0109. Output bills at $0.47 per million and uncached input at $0.15, so log both when a task runs long reasoning, since either side can drive the bill. The API allows up to 131,072 output tokens per call, and one call that uses all of it after a 10,000 token prompt costs $0.06310384. Set an output cap on unattended jobs and read the usage the API returns, so you know whether reasoning tokens are counted in the output total you pay for. The context window is 1,000,000 tokens on the API.

Verdict

Measure the output side first. For a reasoning model, the length of its answers and its thinking decides the bill more than the prompt does, so log real output tokens from a sample of your traffic and price that sample in the calculator. Then look at the cache. Put the instructions and shared context at the start of each request so repeated calls can reuse them. No batch prices are listed right now, so spread work out by capping output and trimming prompts instead. If your volume gets high enough that you'd consider running the open Qwen3.8-Flash-Next weights yourself, price the hardware with the VRAM calculator and put it next to the monthly API figure from this page. Read the Qwen community license before any commercial use, and test that the open weights behave like the API model on your task.