Qwen Token Counter
Estimate your text in Qwen tokens and price it across 5 Alibaba models — all of which hold 1,000,000 tokens, about 1,493 pages. Qwen is where the market floor is: Qwen3.5 Flash at $0.100 per 1M input is the cheapest million-token-window model in this catalogue, on both input and output.
- Countingest.
- Context1M
- Input$2.50
- Output$7.50
- Cached$0.25010% of input
Estimate Qwen tokens
Opens on Qwen3.7 Max, which is the one Qwen model this counter can price against a live count — the other 4 rows below are priced for comparison only.
Drop files here or click to browse
Supports PDF, TXT, MD, JSON, CSV, XML — max 10MB
Count as an API request
Providers also bill the role and boundary tokens that wrap your text. Anthropic's own example — a short system prompt plus one message — costs 14 input tokens for 7 tokens of visible text.
Which Qwen models can this tool count?
Only 1 of the 5 rows below feeds the counter above; the rest are in the table for price comparison only. Alibaba does not publish the tokenizer its Qwen models use, so even that one is an estimate rather than an exact count — calibrated on English prose, and looser on code, JSON and Chinese, where a character often costs close to a whole token.
Two pricing caveats matter more than the count’s precision. Alibaba tiers by request size: crossing an input threshold — 256K tokens on most models, 32K on the coder tier — re-prices the entire request, sometimes several times over. And the rates here are the Singapore international price list; the Beijing region is billed separately and differs. Check both against Alibaba’s own table before you commit a budget.
Qwen models and prices
5 Alibaba models, all with a 1,000,000-token window and all reading cached input at 10% of their input rate. The spread is wide for a single family: Qwen3.7 Max at $2.50 in and $7.50 out down to Qwen3.5 Flash at $0.100 and $0.400 — a factor of 25 on input.
| Model | Inputper 1M tokens | Cachedper 1M tokens | Outputper 1M tokens | Context | Counting | Your cost |
|---|---|---|---|---|---|---|
| Qwen3.7 Max | $2.50 | $0.250 | $7.50 | 1M | est. | — |
| Qwen3.7 Plus* | $0.400 | $0.040 | $1.60 | 1M | est. | — |
| Qwen3.6 Flash* | $0.250 | $0.025 | $1.50 | 1M | est. | — |
| Qwen3.5 Flash* | $0.100 | $0.010 | $0.400 | 1M | est. | — |
| Qwen3 Coder Plus* | $1.00 | $0.100 | $5.00 | 1M | est. | — |
Rates in USD per 1M tokens, read from Alibaba’s own pricing page on August 22, 2026. Providers change prices without notice.
Qwen’s context window
Every Qwen model in this table accepts 1,000,000 tokens — about 1,493 pages of prose at roughly 670 tokens a page. As with Google, the cheap tiers are not penalised on room: Qwen3.5 Flash takes the same size prompt as Qwen3.7 Max, which is what makes the low rates interesting rather than merely low.
The size tiers are the thing to watch. Crossing about 256,000 input tokens re-prices the whole request on most Qwen models, so the top two thirds of that window are not available at the rates shown — on the Plus tier the tiered rate reaches $1.20 input and $4.80 output, and on the coder tier, where the threshold is only 32,000 tokens, $6.00 and $60.00. At the standard rate, sending 200,000 tokens to Qwen3.7 Max costs $0.500, and to Qwen3.5 Flash $0.020.
What Qwen tokens cost
Three request shapes priced on Qwen3.7 Max and Qwen3.5 Flash, at standard-tier Singapore rates. The third crosses Alibaba’s size threshold, so a real bill for it would be higher than shown.
| Request | Qwen3.7 Max | Qwen3.5 Flash |
|---|---|---|
| One chat turn1,000 in / 500 out | $0.0063 | $0.0003 |
| A long document100,000 in / 2,000 out | $0.265 | $0.011 |
| A bulk job1,000,000 in / 50,000 out | $2.88 | $0.120 |
Qwen3.5 Flash is the reason this page is worth reading if you are cost-sensitive. At $0.100 in and $0.400 out it is the cheapest model with a million-token window anywhere in this 56-model catalogue, on both rates — around a fifth of what the nearest large-context alternatives charge for output. Cached reads across the family are a flat 10% of input, in line with OpenAI, Anthropic and Google. Confirm the region and the size tier before you build a forecast on these numbers.
Qwen token counter FAQ
How many tokens is 1,000 words in Qwen?
About 1,330 tokens for English prose, and 1,000 tokens is roughly 752 words. Alibaba publishes no vocabulary, so this is an estimate. Qwen is frequently used on Chinese text, which tokenizes closer to one token per character — the estimate above assumes English, so add a margin for anything else.
Which Qwen models can this tool count?
1 of 5: Qwen3.7 Max is wired into the counter, and the remaining 4 rows appear in the table for price comparison only. That is a deliberate limit rather than an oversight — without a published tokenizer, adding more rows to the picker would imply a precision that does not exist. All 5 are priced from Alibaba’s own table.
What is Qwen’s context window?
1,000,000 tokens across all 5 models here — roughly 1,493 pages of prose. The uniformity is real, but the pricing is not uniform across it: crossing about 256,000 input tokens re-prices the entire request on most Qwen models, so the affordable window is considerably smaller than the technical one.
Is Qwen cheaper than GPT or Claude?
Substantially, at the bottom of its range. Qwen3.5 Flash costs $0.100 per 1M input and $0.400 per 1M output, which makes it the cheapest million-token-window model in this entire 56-model catalogue on both rates. Qwen3.7 Max, the flagship, is $2.50 / $7.50 — comparable to mid-tier GPT and Claude models rather than dramatically below them. As always, the answer depends on your input-to-output ratio, which the calculator above will price for you.
How much does Qwen3.7 Max cost?
$2.50 per 1M input tokens, $7.50 per 1M output and $0.250 for cached reads — 10% of the input rate. Those are standard-tier Singapore figures; a request that crosses Alibaba’s input threshold is re-priced in full, and the Beijing region has its own table.
Does Qwen pricing change by region?
Yes, and it is a common source of surprise. The rates on this page come from Alibaba’s Singapore (international) price list, which is what most non-China users are billed against; the Beijing region is priced separately and differs. Model Studio also applies size tiers that re-price a whole request once the input crosses a threshold. Both caveats are listed with the notes under the table, and both are worth confirming against Alibaba’s own documentation before you plan around a number.
Qwen against the field
Qwen sets the floor for large-context pricing, but regional tables and size tiers mean the headline rate is not always the rate you pay. Compare 56 models from 11 providers with every caveat spelled out, or open another brand’s counter below.
New to tokens? Read how tokenization works