Qwen3 Coder Pricing (2026)

Qwen3-Coder is Alibaba's coding-specialist tier, and measured via OpenRouter it routes at $0.22 per million input tokens and $1.80 output, served as qwen3-coder-480b-a35b; in our own test it was the fastest and cheapest coding call we ran. That makes it the tier most worth checking before you default to a general model for code generation.

Two prices follow this model around, and conflating them is the most common budgeting mistake we see. The OpenRouter-routed rate is what you pay if you reach Qwen through that marketplace, and it is the number we verified live on 2026-06-26. Alibaba's own endpoint quotes a different figure. We keep them on separate lines below because they are not interchangeable, and one of them still needs native re-verification.

Qwen3-Coder API pricing (verified 2026-06)

Qwen3-Coder is a code-specialized large language model that Alibaba serves under the internal name qwen3-coder-480b-a35b, a mixture-of-experts checkpoint that activates roughly 35 billion of its 480 billion parameters per token. The practical upshot of that architecture is the price point: you get a frontier-class coding model billed like a mid-tier one.

According to [OpenRouter's published catalog as we read it on 2026-06-26], the routed rate is $0.22 input and $1.80 output per million tokens. Note that this is the marketplace-routed price, not Alibaba's list price. According to Alibaba Cloud Model Studio, the official Qwen3-Coder rate sits near $0.15 input and $0.60 output per million tokens, though it varies by SKU and variant and we have flagged it for native re-verification. The gap is real and it cuts against the usual assumption: on output tokens, the official endpoint can be cheaper than the marketplace, which matters because code generation is output-heavy.

Source	Input ($/1M)	Output ($/1M)	Disclosure
OpenRouter-routed (verified)	$0.22	$1.80	served as qwen3-coder-480b-a35b
Alibaba Model Studio (official)	~$0.15	~$0.60	variant-dependent, needs native re-verify
DeepSeek V4-Flash (anchor)	$0.14	$0.28	official api.deepseek.com

The table makes the routing decision concrete. If your workload is output-heavy, the official Alibaba endpoint is the cheaper home for Qwen3-Coder once you have native access confirmed; OpenRouter buys you convenience at an output premium.

We tested Qwen3-Coder on a live coding prompt (first-hand evidence)

Published rates tell you the per-token math. They do not tell you what a real coding turn costs or how fast it lands, so we called the endpoint and measured both. The full run sits in the cluster evidence pack; here is the coding call.

We sent a coding prompt to Qwen3-Coder via OpenRouter and the request resolved to qwen3-coder-480b-a35b-07-25. It consumed 36 input tokens, returned 58 completion tokens, billed $0.0000656, and completed in 1.50 seconds. That was the fastest and cheapest single call across every Qwen tier we exercised that day, faster than qwen3-max on the same coding task and a fraction of the cost.

The reason that number matters is per-task economics, not headline rate. A sub-second-and-a-half coding turn at well under a hundredth of a cent means an interactive coding assistant running thousands of small completions stays trivially cheap on the model line; latency, not price, becomes your constraint. We re-read the artifact to be sure the served model string was the 480b-a35b checkpoint and not a fallback, because OpenRouter routing can substitute, and on this run it did not.

For teams that want to skip the marketplace markup, native access exists. According to Alibaba Cloud Model Studio, Qwen models are reachable to US and international developers through the Model Studio International endpoints, including a Singapore region with a free starter quota, which is the route to confirm before you trust the lower official output rate in your own budget.

One caution from the same test session. The sibling qwen3.5-flash model, despite the reassuring name, produced 2,831 output tokens on a two-sentence prompt and took 17.85 seconds, making it the slowest and most expensive call in our set. Do not assume a "flash" label means fast or cheap; budget against measured behavior, not the marketing name.

Per-task coding cost versus DeepSeek V4-Flash

The honest comparison for a coding budget is per-task, not per-million, because token counts differ wildly by model verbosity. We anchored Qwen3-Coder against DeepSeek V4-Flash, the cheapest credible coding tier we track.

For the DeepSeek anchor we called its official endpoint directly. According to DeepSeek's pricing docs, V4-Flash bills $0.14 input and $0.28 output per million tokens, and on a coding prompt the run used 27 input and 120 output tokens at 3.22 seconds. Worth stating plainly: the comparison is asymmetric. We reached Qwen via OpenRouter and DeepSeek on its native API, so the Qwen figure carries a marketplace markup the DeepSeek figure does not.

Model	Route	Coding-run tokens (in/out)	Latency	Per-token rate (in/out)
Qwen3-Coder	OpenRouter	36 / 58	1.50s	$0.22 / $1.80
DeepSeek V4-Flash	official	27 / 120	3.22s	$0.14 / $0.28

On raw output rate DeepSeek V4-Flash is far cheaper, $0.28 against $1.80 per million. But Qwen3-Coder was less verbose on the coding task (58 output tokens versus 120) and returned in less than half the time, so for short, frequent completions the latency edge can outweigh the rate gap. For long code generation where output volume dominates, DeepSeek's output rate wins decisively. For deeper DeepSeek economics see the DeepSeek API pricing hub.

Choose Qwen3-Coder if latency and code quality lead your shortlist

Pick Qwen3-Coder when you want a coding-specialist model with low latency and you value the 480b-a35b checkpoint's code competence over the absolute cheapest token. If you route through OpenRouter, accept the $1.80 output rate as the cost of convenience, and prefer Alibaba's official endpoint once native access is confirmed because its output rate is materially lower. Avoid defaulting to it for long-form generation where a cheaper output rate like DeepSeek V4-Flash's compounds in your favor.

FAQ

How much does Qwen3-Coder cost per million tokens? Measured via OpenRouter on 2026-06-26 it routes at $0.22 input and $1.80 output per million tokens, served as qwen3-coder-480b-a35b. Alibaba's official Model Studio rate is lower, near $0.15 input and $0.60 output, but varies by variant and is pending native re-verification.

What did a real Qwen3-Coder coding call cost in your test? On our OpenRouter-routed run the model used 36 input and 58 output tokens, billed $0.0000656, and returned in 1.50 seconds. It was the fastest and cheapest call across every Qwen tier we tested that session.

Is Qwen3-Coder cheaper than DeepSeek V4-Flash? On raw output rate, no. DeepSeek V4-Flash bills $0.28 output per million versus Qwen3-Coder's $1.80 OpenRouter rate. But Qwen3-Coder was less verbose and roughly twice as fast on our coding prompt, so for short frequent completions the per-task gap narrows.

Why are there two different prices for Qwen3-Coder? One is the OpenRouter-routed marketplace price we verified live; the other is Alibaba's official Model Studio list price. They are not the same number and the official figure still needs re-checking on a native key, so we never present the routed price as Alibaba's official rate.

This is part of the Qwen API pricing hub, which compares every Qwen tier alongside DeepSeek and Western anchors.

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

Qwen3-Coder is Alibaba's coding-specialist tier, and measured via OpenRouter it routes at $0.22 per million input tokens and $1.80 output, served as…

Written by Fan Chuanyu

Qwen3-Coder API pricing (verified 2026-06)

We tested Qwen3-Coder on a live coding prompt (first-hand evidence)

Per-task coding cost versus DeepSeek V4-Flash

Choose Qwen3-Coder if latency and code quality lead your shortlist

FAQ