Qwen API pricing: 51 paid models per 1M tokens

According to APITariff, Qwen API pricing on OpenRouter on 2026-10-03 has its lowest paid row at qwen3.7-flash, the lowest of the 51 paid Qwen text rows by input price, output price and cost per call: $0.03 input and $0.13 output per 1M tokens. On the standard workload used on this page, 1,000 input and 300 output tokens per call, that is $69 per 1,000,000 calls, against $3,800 for qwen3.8-max, 55 times more. Alibaba Cloud’s Model Studio page, marked “Last Updated: Oct 03, 2026”, lists the same $0.03 and $0.13 for qwen3.7-flash on requests up to 32K tokens, and $2 and $6 for qwen3.8-max in its Singapore region.

Of 8 models we compared across both sources, 4 carry the identical price, 1 matches Alibaba’s 20% limited-time discount, and 3 sit 35% to 41% below Alibaba’s list price on OpenRouter. OpenRouter listed 52 Qwen text slugs in our fetch of 2026-10-03: 51 paid and 1 at $0.

Last updated 2026-10-05 · 52 Qwen text slugs on OpenRouter · source: Alibaba Cloud Model Studio pricing page and APITariff fetch of OpenRouter, 2026-10-03; APITariff fetch of kie.ai, 2026-10-05. Rates move: 10 of 57 Qwen slugs changed price in our 23 fetches since 2026-09-02, so recheck before a high-volume commitment.

Qwen API pricing per 1M tokens on Alibaba Cloud and OpenRouter

Eight Qwen text models are priced below from Alibaba’s Singapore (International) list and from OpenRouter, both read on 2026-10-03.

Model	Alibaba, input / output	OpenRouter, input / output	OpenRouter context
qwen3.8-max	2 / 6	2.00 / 6.00	1,000,000
qwen3.7-max	2.5 / 7.5	1.475 / 4.425	1,000,000
qwen3.7-plus	0.4 / 1.6 list, 20% off	0.32 / 1.28	1,000,000
qwen3.8-flash	0.15 / 0.47	0.15 / 0.47	1,000,000
qwen3.7-flash	0.03 / 0.13	0.03 / 0.13	1,000,000
qwen3-coder-plus	1 / 5	0.65 / 3.25	1,000,000
qwen3-max	1.2 / 6	0.78 / 3.9	262,144
qwen3.8-2.4t-a95b	2 / 6	2.00 / 6.00	1,048,576

8 of 52 OpenRouter Qwen text slugs shown. Input / output in $ per 1M tokens. Sources: Alibaba Cloud Model Studio pricing, Singapore region, deployment scope International, page marked “Last Updated: Oct 03, 2026”; APITariff fetch of openrouter.ai/models, 2026-10-03 08:40 UTC. Tiered Alibaba models show the first tier. The OpenRouter qwen3.8-max slug is qwen3.8-max-0902.

Alibaba bills by the size of each request, so qwen3.7-flash (3 tiers), qwen3-max (3 tiers) and qwen3-coder-plus (4 tiers) have higher rates than the row shown. The tier section below prices that jump.

How much does the Qwen API cost per month?

On OpenRouter prices of 2026-10-03, 100,000 calls a month of the standard workload cost $6.90 on qwen3.7-flash and $380 on qwen3.8-max-0902. The 7 rows below, picked to span the price range and sorted by cost at 1M calls, run from $6.90 to $760 at that volume.

OpenRouter slug	10k calls	100k calls	1M calls
qwen3.7-flash	$0.69	$6.90	$69
qwen3.8-flash	$2.91	$29.10	$291
qwen3.7-plus	$7.04	$70.40	$704
qwen3-coder-plus	$16.25	$162.50	$1,625
qwen3.7-max	$28.03	$280.25	$2,802.50
qwen3.8-max-0902	$38	$380	$3,800
qwen3.8-max-prime	$76	$760	$7,600

7 of 51 paid Qwen rows shown. Computed from the OpenRouter column of the first table and the qwen3.8-max-prime slug at $4.00 input and $12.00 output. Standard workload: 1,000 input plus 300 output tokens per call.

What does the Qwen 3.8 Max API cost?

On 2026-10-03, Alibaba’s Singapore list shows qwen3.8-max at $2 input and $6 output per 1M tokens, in one tier for requests up to 1M tokens. The China (Beijing) and Global deployment rows on the same page list $1.65 and $4.951, so the region you call decides the rate.

OpenRouter lists qwen3.8-max-0902 at $2.00 and $6.00, equal to Alibaba Singapore, and a separate qwen3.8-max-prime slug at $4.00 and $12.00, twice that. Alibaba’s page prices the prime variant at $3.301 and $9.902 in its Beijing section and points to a “Fast mode (Prime)” page. On the standard workload at 1,000,000 calls the two OpenRouter slugs bill $3,800 and $7,600.

Is the Qwen API free?

Partly. On 2026-10-03, Alibaba’s Singapore rows list a free quota of 1 million tokens per model, “Valid for 90 days from the date of Model Studio activation, model release, or application approval, whichever is later”. After that the rates in the first table apply.

OpenRouter listed 1 Qwen text slug at $0 on 2026-10-03, qwen3.8-27b:free, against 51 paid slugs. No qwen3.7 slug is free: qwen3.7-flash lists $0.03 and $0.13, qwen3.7-plus $0.32 and $1.28, and qwen3.7-max $1.475 and $4.425. The caps we counted on OpenRouter’s free models on 2026-09-29 are in Free AI API: 50 requests a day, or 1,000 after $10 of credit. We did not fetch the cap for the qwen3.8-27b:free slug itself, so treat it as unconfirmed for that slug.

Is Qwen cheaper than DeepSeek?

At the floor, no: on cost per call alone, deepseek-v4-flash is the lower row. On OpenRouter on 2026-10-03, deepseek-v4-flash costs $44.80 per 1,000,000 calls of the standard workload ($0.028 input, $0.056 output), and the lowest Qwen row, qwen3.7-flash, costs $69, which is 1.5 times as much. None of the 51 paid Qwen rows costs less than deepseek-v4-flash.

Higher up the list the gap closes: deepseek-v4-pro costs $334.08 on the same workload ($0.2088 and $0.4176), and 16 of the 51 paid Qwen rows cost less than that. Qwen lists 51 paid text slugs against 12 for DeepSeek, so a count of rows under a price says more about family size than about price level. The DeepSeek API pricing guide carries tables dated 2026-09-17, and its OpenRouter column shows deepseek-v4-pro-0813 at $1.32 input, against $0.2088 on the current deepseek-v4-pro slug. Neither price list says anything about output quality, and we did not bench either family.

Why do Alibaba and OpenRouter prices differ?

We do not know. The readings are these, all on 2026-10-03: 4 of 8 compared models carry Alibaba’s list price on OpenRouter, qwen3.7-plus carries Alibaba’s discounted price, and 3 sit under it. qwen3.7-max is 41% under ($1.475 against $2.50 input), and qwen3-coder-plus and qwen3-max are each 35% under Alibaba’s 0 to 32K tier.

For a budget ceiling, use Alibaba’s list. On the standard workload at 1,000,000 calls, qwen3.7-max bills $4,750 at Alibaba’s list and $2,802.50 on OpenRouter, a gap of $1,947.50. Alibaba’s page says it “only lists standard prices” and sends readers to the console for promotions.

What is the tier cliff on Qwen models?

Alibaba prices each request by its total input size, and “all tokens in the request are billed at the unit price of the corresponding tier”. On Alibaba’s list read 2026-10-03, qwen3.7-flash bills $0.03 input and $0.13 output up to 32K tokens, $0.10 and $0.40 from 32K to 256K, and $0.20 and $0.80 from 256K to 1M.

A request of 100,000 input tokens therefore bills all of them at $0.10 per 1M: $0.0100, against $0.0030 at the first-tier rate, a factor of 3.33. OpenRouter prints one price per slug, $0.03 and $0.13 for qwen3.7-flash, and we did not test which tier it applies to long requests. Until that is checked, price requests above 32K tokens at Alibaba’s higher tier rates.

How often do Qwen prices change on OpenRouter?

We fetched OpenRouter’s model list 23 times between 2026-09-02 and 2026-10-03 and saw 57 Qwen text slugs in total, of which 52 were listed on the last date. 10 of the 57 changed price at least once; the counts below are observed changes, not a trend.

OpenRouter slug	Readings, input / output (fetch date)
qwen3.8-27b	0.425 / 2.55 (09-02), 0.42 / 3.0 (09-08), 0.214 / 2.55 (09-15), 0.2 / 2.5 (09-21), 0.0749 / 4.4 (09-29)
qwen3.6-27b	0.6 / 3.6 (09-02), 0.3 / 2.0 (09-08), 0.32 / 2.7 (09-23), 0.32 / 3.2 (09-26)
qwen3.5-35b-a3b	0.25 / 1.25 (09-02), 0.3125 / 1.25 (09-08), 0.1625 / 1.3 (09-15), 0.15 / 1.0 (10-02)

3 of 10 changed slugs shown. $ per 1M tokens, a reading is listed only when the price differs from the one before it. Source: APITariff fetch of openrouter.ai/models. The qwen3.8-27b paid slug is absent from the 2026-10-03 fetch, where only qwen3.8-27b:free is listed.

Input and output moved in opposite directions on qwen3.8-27b between 09-21 and 09-29: input fell from $0.2 to $0.0749 while output rose from $2.5 to $4.4. The output price rose 76% (4.4 / 2.5 = 1.76) in the 8 days between those two readings.

Does kie.ai list the Qwen API?

No. kie.ai, a reseller, listed 42 text models in our fetch of 2026-10-05, and none of them is a Qwen text model. Its lowest text row on that date is gpt-6-luna at $0.03 input and $0.15 output per 1M tokens. On the standard workload that bills $75 per 1,000,000 calls, $6 or 8.7% more than qwen3.7-flash at $69 on OpenRouter.

The gpt-6-luna price was the same in all 11 of our kie.ai fetches since 2026-09-25, and the 9 OpenRouter Qwen slugs in the two price tables above read the same on 2026-10-05 as on 2026-10-03. This is a price reading only: gpt-6-luna is a different model family, and we did not bench it against any Qwen model.

kie.ai does list Qwen image models: 22 rows in the same fetch, for example Qwen image 3.0 text to image at $0.024 per image. This page prices text tokens only, so those rows are not compared here. The kie.ai price page lists every kie.ai row we hold.

What this page does not cover

Not covered: the Alibaba regions other than Singapore International, open-weight self-hosting cost, resellers other than OpenRouter and the kie.ai check above, thinking-mode output surcharges beyond the rows quoted, and any measure of speed or answer quality. Every rate here was read on 2026-10-03, and the 10 changed slugs above show that OpenRouter moves rates within days.

Open the calculator with 1,000 input and 300 output tokens to price any of the 52 rows, and see the LLM API price ranking for where qwen3.7-flash sits among all text models. The methodology page states how each fetch works.

Cite as: APITariff, apitariff.com/guides/qwen-api-pricing/, read 2026-10-03.

Drafted with AI from APITariff’s own price data, checked and edited by a person. The chart is rendered from the tables above, not generated.

Change log

• 2026-10-03First edition. Alibaba Singapore list and OpenRouter, both read 2026-10-03.
• 2026-10-05Added the kie.ai check: no Qwen text model among 42 kie.ai text rows fetched 2026-10-05. Price tables unchanged; the 9 OpenRouter slugs in them read the same on 2026-10-05.