Cheapest Qwen Model (2026)

The cheapest Qwen model by per-token rate is qwen3.5-flash at $0.065 input / $0.26 output per million tokens via OpenRouter, but our own test showed it can cost more per task than qwen3-max once a "flash" model decides to ramble. That gap is the part most buyers miss, so this ranking pairs the rate card with what we actually paid per call.

If you only read one line, read this: the cheapest sticker price and the cheapest finished task are not the same number, and on Qwen the difference is large enough to flip the ranking.

The cheapest Qwen API models, ranked (verified 2026-06)

All six prices below are OpenRouter-routed values we read from the live OpenRouter catalog on 2026-06-26. They are not Alibaba's official rate card; see the official-pricing note further down. Per million tokens, USD, cheapest input first.

Rank	Model (via OpenRouter)	Input $/1M	Output $/1M	Best fit
1	qwen3.5-flash	$0.065	$0.26	High-volume short replies
2	qwen3-32b	$0.08	$0.28	Cheap general chat
3	qwen3-235b-a22b-2507	$0.09	$0.10	Flagship-class, output-heavy
4	qwen-plus	$0.26	$0.78	Balanced general workloads
5	qwen3-coder	$0.22	$1.80	Code generation
6	qwen3-max	$0.78	$3.90	Hardest reasoning

The ranking above sorts by input price, which is the conventional way buyers shop. But look at the third row. qwen3-235b-a22b-2507 charges $0.09 in and just $0.10 out, the flattest input-to-output ratio in the table. For any workload that generates far more tokens than it reads, that model is effectively cheaper than the two rows above it, because output is where most chat bills are won or lost.

What surprised us: the "flash" model was the most expensive call we made

A flash model is a tier vendors position as the fast, cheap option for high-volume short tasks, which usually means a small or distilled model tuned for low latency. The name sets an expectation. Our test broke it.

When we called the endpoint on 2026-06-26 (measured via OpenRouter), qwen3.5-flash answered a two-sentence prompt with 2,831 output tokens, taking 17.85 seconds and billing $0.00073866 for that single call. That made it both the slowest and the most expensive run in our entire Qwen set, ahead of qwen3-max, the supposed premium tier. We re-checked the artifact to be sure the token count was real and not a logging glitch; it was real.

Compare that to qwen3-max on a general prompt, which returned 63 output tokens in 3.22 seconds for $0.00027534, also measured via OpenRouter. The flash model's per-token rate is roughly twelve times cheaper than max, yet it cost nearly three times as much to finish one task. The lever was not the rate. It was output length, and flash chose to write an essay.

The practical upshot: a low per-token price only translates into a low bill if you cap output. Set max_tokens, and the per-token ranking holds. Leave it open, and a verbose model at $0.26/1M can out-bill a premium model at $3.90/1M on the same question.

First-hand cost per call (measured via OpenRouter, 2026-06-26)

Documentation gives you the rate card. It does not tell you what a real call costs, because that depends on how many tokens the model decides to emit. Here is what we billed, one call each, routed through OpenRouter.

Model	Prompt tok	Output tok	Latency	Billed cost
qwen3-coder	36	58	1.50s	$0.0000656
qwen3-32b	39	120	3.61s	$0.00003672
qwen-plus	38	94	2.47s	$0.0000832
qwen3-max (coding)	36	53	2.28s	$0.00023478
qwen3.5-flash	40	2,831	17.85s	$0.00073866

The cheapest finished call in the set was qwen3-coder at $0.0000656, served as qwen3-coder-480b-a35b and also the fastest at 1.50 seconds. That inverts the sticker ranking, where qwen3-coder sits fifth. On short tasks where output stays disciplined, the model with the higher headline rate but tighter responses can be the cheaper one to run.

Official Alibaba pricing is a separate, higher number

The prices above are what a third-party router charges. Alibaba's own endpoint is a different story, and conflating the two is the most common pricing mistake we see.

According to Alibaba Cloud Model Studio pricing, Qwen3-Max on the International endpoint is tiered by input size and runs roughly $2.40 input / $12.00 output per million tokens for the 32K-128K bracket. That is materially higher than the $0.78 / $3.90 we saw routed through OpenRouter. The buyer insight hiding in that gap: for the flagship, an international developer can pay less through a router than through Alibaba's own International endpoint. We flag this as pending native re-verification, since we measured the routed price live but read the official figure from sourced research, not from a native DashScope key.

According to Alibaba Cloud Model Studio pricing, US and international developers reach Qwen through Model Studio International, with a Singapore endpoint offering roughly 1M free tokens for 90 days and a US (Virginia) deployment with no free quota. So access is not the blocker; price-source confusion is. Always confirm which endpoint a quoted Qwen price came from before you budget against it.

How Qwen's cheapest tier stacks up against DeepSeek

A neutral comparison needs a second house. According to DeepSeek API Docs, DeepSeek V4-Flash bills $0.14 input / $0.28 output per million tokens at its official endpoint. Our DeepSeek anchor call (27 in / 120 out, 3.22s on the official api.deepseek.com endpoint) cost is computed from that official rate, not billed, so note the asymmetry: DeepSeek ran on its native endpoint while every Qwen number here is OpenRouter-routed.

On input, qwen3.5-flash ($0.065) and qwen3-32b ($0.08) both undercut DeepSeek V4-Flash ($0.14). On output, qwen3-235b-a22b-2507 at $0.10 is the cheapest token-generator in either lineup. For a fuller side-by-side across providers, the DeepSeek API pricing breakdown covers that house in depth. The honest verdict: Qwen wins the headline cheap-input race, but DeepSeek's flatter, single-endpoint pricing is easier to budget against, and our flash-model result is a warning that headline cheapness needs a max_tokens guardrail before it means anything.

Verdict: cheapest is not always cheapest per task

Choose qwen3.5-flash or qwen3-32b if your workload is high-volume and your outputs are short and capped. Choose qwen3-235b-a22b-2507 if you generate far more than you read, because its $0.10 output rate is the real saver. Reach for qwen3-max only when the task needs the reasoning, and accept the $3.90 output. Above all, set max_tokens before you trust any per-token ranking, because a flash model with no ceiling can out-bill the flagship on a single question.

FAQ

What is the cheapest Qwen API model? By per-token input rate, qwen3.5-flash at $0.065/1M input and $0.26/1M output via OpenRouter (measured 2026-06-26). But on a per-finished-task basis it was the most expensive call in our test, because it emitted 2,831 output tokens on a two-sentence prompt, so cap max_tokens before relying on its low rate.

Is qwen3.5-flash really the cheapest to run? Not always. Its per-token price is the lowest, but our live OpenRouter call billed $0.00073866 for one verbose response, more than qwen3-max's $0.00027534 on a similar prompt. Cheapest-per-token and cheapest-per-task diverge whenever output length is uncapped.

Are these Qwen prices Alibaba's official rates? No. Every Qwen price here is OpenRouter-routed and disclosed as such. Alibaba's official Model Studio Qwen3-Max International pricing runs higher, around $2.40/$12.00 per million for the 32K-128K bracket, and is pending native re-verification on a DashScope key.

Can US developers access Qwen? Yes. Per Alibaba Cloud Model Studio, US and international developers use Model Studio International, with a Singapore endpoint (about 1M free tokens for 90 days) and a US Virginia deployment with no free quota.

How does the cheapest Qwen compare to DeepSeek? On input, qwen3.5-flash ($0.065) and qwen3-32b ($0.08) undercut DeepSeek V4-Flash's $0.14. On output, qwen3-235b-a22b-2507 at $0.10 beats DeepSeek's $0.28. DeepSeek's single official endpoint is simpler to budget, while Qwen's cheapest routed prices come with a max_tokens caveat.

This is part of the Qwen API pricing hub, which ranks every Qwen tier and routing path in one place.

Author: Kevin Fan, Customer Success Manager at China LLM Directory, specializing in Chinese LLM ecosystem pricing. Last verified: 2026-06-26.

The cheapest Qwen model by per-token rate is qwen3.5-flash at $0.065 input / $0.26 output per million tokens via OpenRouter, but our own test showed it…

Written by Fan Chuanyu