Qwen Image 2.0 API: Integration Guide, Pricing & Code

Yes — the Qwen Image 2.0 API is real, documented, and callable today. Alibaba Cloud Model Studio still lists qwen-image-2.0 and qwen-image-2.0-pro, still prices them per successfully generated image ($0.035 and $0.075 at the Singapore region as of September 6, 2026), and still serves them from the same DashScope multimodal-generation endpoint used by the newer models. What changed since Qwen Image 3.0 launched in July 2026 is the positioning, not the availability: 2.0 is now the previous generation — frozen but officially served — and this guide tells you exactly where that leaves an integration, with every number re-pulled from the official pages today.

One disclosure before the code: Modellix is an AI model API aggregator and has a commercial interest in this article. Qwen Image 2.0 is not in Modellix’s catalog as of September 6, 2026 — its model page renders “Model Not Found,” so there is nothing live behind that URL to route you to or price. The routing section shows you the honest current option on our side (the Qwen Image 3.0 family) while everything before it stays vendor-neutral integration material you can use with or without us.

What Qwen Image 2.0 is in 2026: the previous generation, still officially served

Qwen-Image-2.0 was announced in February 2026 (blog February 7; weights and API rollout February 10) as a single 7B-class model that merged the generation and editing tracks Alibaba had run separately through 2025 — native 2K output, professional typography rendering driven by prompts up to roughly 1,300 tokens, and unified image generation and editing in one call. The official release post frames it around infographics, posters, comics, and multi-image fusion, which is still the family’s signature strength: legible text inside the generated image.

The API exposes the generation in two quality tiers, plus dated snapshot IDs that trip up everyone who reads the docs casually:

Model ID (what you send)	Tier	Notes
qwen-image-2.0	Accelerated	Balances quality and speed; currently equivalent to the qwen-image-2.0-2026-03-03 snapshot
qwen-image-2.0-pro	Pro	Stronger text rendering, realism, and semantic adherence; currently equivalent to the qwen-image-2.0-pro-2026-04-22 snapshot
qwen-image-2.0-2026-03-03, qwen-image-2.0-pro-2026-03-03/-2026-04-22/-2026-06-22	Pinned snapshots	Explicit dated versions you can call directly for reproducibility

The snapshot mechanism is the part most third-party pages skip: Model Studio pins each model name to a dated release (-2026-03-03, -2026-04-22, and, for Pro, -2026-06-22), and the bare alias tracks one of them. Call the bare ID and you get the alias’s current snapshot; call a dated ID and your output won’t silently change when Alibaba updates the alias. The last 2.0-series snapshot is dated June 2026 — since Qwen Image 3.0 was announced on July 20–21, 2026, no new 2.0 snapshots have shipped. That is the clearest available signal of its status: not deprecated, not removed, just no longer being developed — Alibaba’s own pricing and API pages still list it without any legacy label as of today.

The Qwen-Image generation ladder: 2.0 sits one rung below the current 3.0, still served but no longer updated. Diagram generated September 6, 2026 — no prices, because they drift.

For the full family picture — where 2.0 sits against qwen-image, plus, max, the edit variants, and the current 3.0 lineup — our Qwen Image API guide is the wider map; this article is the 2.0 version-specific deep dive.

Qwen Image 2.0 API pricing, pulled from Model Studio today

Official pricing is per successfully generated image, billed on the Alibaba Cloud Model Studio page — if a call returns three images out of four requested, you pay for three. The rates below were read from the official model pricing page on September 6, 2026, Singapore (International) region — the region to use without a Chinese mainland account:

Model ID	Official price (Singapore, per image)	Free quota
qwen-image-2.0	$0.035	100 images
qwen-image-2.0-2026-03-03	$0.035	100 images
qwen-image-2.0-pro	$0.075	100 images
qwen-image-2.0-pro-2026-03-03 / -2026-04-22 / -2026-06-22	$0.075	100 images

Two notes on reading that table. First, the free quota and those prices apply only in the Singapore region; Beijing is a separate deployment with its own prices and, per the same page, no free quota, while Hong Kong, Frankfurt, and Tokyo do not host the 2.0 series on the official page. Second, these are the text-to-image rates — because 2.0 unifies generation and editing, the same model IDs also appear in the image-editing section at identical prices, so an edit call costs the same $0.035/$0.075 as a generation.

For budget math: a single qwen-image-2.0 image costs $0.035, so a 6-image batch (n: 6) costs at most $0.21 — and usually less, because only successful images are billed. The same batch on qwen-image-2.0-pro costs at most $0.45. At 1,000 successful images on the standard tier you are looking at $35 before any quota. For the provider-wide pricing context across every Alibaba model, our Alibaba API pricing guide has the full table.

Getting a Qwen Image 2.0 API key: regions first, then the console

You need a Model Studio account and an API key, and the single most common setup failure is mixing regions. Beijing and Singapore are separate deployments with separate keys and separate endpoints — a Beijing key against a Singapore endpoint fails with authentication or service errors, and the console makes the split easy to miss because both regions look identical.

1. Sign in to the Model Studio console and open API Keys (the Singapore console if you are outside mainland China).
2. Create a new key — it starts with sk- and is shown once, so copy it immediately.
3. Find your Workspace ID in the console; the endpoints below embed it as {WorkspaceId}.
4. Keep that key and endpoint pair together for the life of the integration.

The official Qwen-Image API reference documents the full setup and parameter list. Region-scoped keys are the norm across Model Studio, not a quirk of 2.0 — the same rule applies to the 3.0 series and to the text models.

Calling the Qwen Image 2.0 API: curl and Python code

The 2.0 series runs on the synchronous multimodal-generation endpoint — one request, one response, no polling. (The async task workflow with the X-DashScope-Async: enable header exists only for the older qwen-image and qwen-image-plus models.) The endpoint, from the official reference:

Region	Endpoint
Singapore	https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation
Beijing	https://{WorkspaceId}.cn-beijing.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation

curl — note the request shape: model at the top level, the prompt inside input.messages (exactly one user message), and generation controls under parameters:

curl --location 'https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation' \
--header 'Content-Type: application/json' \
--header 'Authorization: Bearer sk-***' \
--data '{
    "model": "qwen-image-2.0-pro",
    "input": {
      "messages": [
        {
          "role": "user",
          "content": [
            { "text": "A dark editorial poster for a developer conference, title text \\\"Build With Qwen Image 2.0\\\" in bold white sans-serif, amber and cyan gradient background, subtle circuit-board line art." }
          ]
        }
      ]
    },
    "parameters": {
      "negative_prompt": "Low resolution, distorted text, warped faces, oversaturated colors",
      "prompt_extend": true,
      "watermark": false,
      "size": "2048*2048"
    }
}'

Python with the DashScope SDK — install dashscope, point it at your region’s base URL, and call MultiModalConversation:

import os
import dashscope

dashscope.base_http_api_url = "https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1"  # Singapore

response = dashscope.MultiModalConversation.call(
    api_key=os.getenv("DASHSCOPE_API_KEY"),
    model="qwen-image-2.0-pro",
    messages=[{"role": "user", "content": [{"text": "A developer-conference poster titled Build With Qwen Image 2.0, bold white sans-serif on an amber and cyan gradient, subtle circuit-board line art"}]}],
    parameters={"size": "2048*2048", "negative_prompt": "Low resolution, distorted text"},
    result_format="message",
    stream=False,
)

if response.status_code == 200:
    image_url = response.output.choices[0].message.content[0]["image"]
    print(image_url)  # download immediately — the URL expires in 24 hours

Two things to know about the response: the image comes back as a URL, not bytes, and that URL — per the API reference — expires after 24 hours and is then purged. Download to your own storage at generation time, not when you deploy. The success payload also reports usage.width, usage.height, and usage.image_count (successful images only), which is what you should bill against internally. The Qwen Image 2.0 API is not OpenAI-compatible at this endpoint — it uses the DashScope multimodal-generation contract, so adapters that expect POST /v1/images/generations will not work without a translation layer.

The 2.0 call is synchronous: send prompt + parameters, receive a temporary PNG URL, download it within 24 hours. Diagram generated September 6, 2026.

Qwen Image 2.0 parameters that actually matter

The parameter set is small, and five knobs change results more than the rest (all documented in the official API reference linked above):

• size — any width*height whose total pixels land between 512×512 and 2048×2048; default 2048*2048. Presets: 2688*1536 (16:9), 1536*2688 (9:16), 2368*1728 (4:3), 1728*2368 (3:4). The older max/plus tiers are locked to smaller preset sets — one more reason 2.0 is the flexible choice.
• n — 1–6 images per call on the 2.0 series; max/plus are fixed at 1 (any other value errors).
• negative_prompt — up to 500 characters of what to exclude (“low resolution, distorted limbs, warped text…”). Truncated silently past 500.
• prompt_extend — defaults to true and silently rewrites your prompt with more detail. Leave it on for diverse output; turn it off for tight composition control.
• watermark and seed — watermark: false by default (no Qwen-Image mark); seed accepts 0–2147483647 and gives similar (not identical) outputs on repeat — treat it as a reproducibility hint, not a guarantee.

Prompt length caps at 1,300 tokens on the 2.0 series (800 on most other Qwen image models), and excess is silently truncated — if a detailed poster brief keeps losing elements, check the cap before blaming the model. Also note the single-turn rule: messages must contain exactly one user message; multi-turn conversations are not supported.

Errors come back in the response body as code + message (HTTP status can still be 200), per the Model Studio error reference. The four integrators actually hit: region mismatch (Beijing key + Singapore endpoint — auth or service errors), InvalidParameter (e.g., num_images_per_prompt must be 1 when you set n > 1 on a model that forbids it), silent truncation (prompt or negative prompt over its cap), and dead image URLs (the 24-hour expiry — download eagerly). None of these are 2.0-specific failures; they are the family’s standard failure modes.

Is there a free Qwen Image 2.0 API?

A limited one, and only in Singapore. Model Studio grants a free quota of 100 images for qwen-image-2.0 and qwen-image-2.0-pro (Singapore region only — other regions have no free quota), valid for 90 days from account activation, per the free quota rules and the pricing page’s quota column. Failed generations are not billed, so they do not burn your quota the way successful ones do. Third-party platforms that host Qwen Image 2.0 (fal.ai, getimg, and others surface on this keyword) run their own trial offers — those are separate from Alibaba’s quota, and their terms change; check each platform rather than assuming parity. For the wider free-tier picture across image APIs, our text-to-image API guide compares the options.

Qwen Image 2.0 vs Qwen Image 3.0: should you upgrade?

Qwen Image 3.0 (announced July 20–21, 2026, per the official announcement) is what Alibaba now points new integrations at. The differences that matter for an API decision:

Dimension	Qwen Image 2.0 (2.0 / 2.0-pro)	Qwen Image 3.0 (3.0 / 3.0-pro)
Prompt capacity	~1,300 tokens	~4,500 tokens (dense layouts: newspapers, exam sheets, storyboards)
Text rendering	Strong, generation-leading for its time	~10px small text, native rendering across 12 languages
Size control	Free w×h, 512²–2048² px, default 2048²	Free w×h within the same pixel band (1K/2K pricing tiers)
Images per call	1–6	1–6
Billing	Flat per-image output price	Input charge (~$0.003/image) + output price that varies by resolution
Snapshot cadence	Frozen since June 2026	Current; new snapshots expected

Pricing on the official Singapore table, pulled today: 3.0 runs $0.03 per image (1K or 2K) plus a $0.003/image input charge; 3.0-pro is $0.04 at 1K and $0.075 at 2K, plus the same input charge — so at 2K the Pro tiers of both generations cost the same per output image, while 3.0 standard is roughly at parity with 2.0 standard once you add the input fee. See the model pricing page linked above for the current table.

When does staying on 2.0 make sense? Three cases, honestly stated: your pipeline already targets it and your prompts are tuned against its quirks (switching models is a regression-testing project, not a find-and-replace); you need a flat per-image price with no input-image surcharge and no resolution-dependent rate; or your workload is batch generation of posters and product visuals where the 3.0 long-prompt advantage (dense multi-panel layouts, small multilingual text) never comes into play. When it does not make sense: any new integration doing infographics, reports, or localized layouts — that is 3.0 territory, and Alibaba is not going to ship new 2.0 snapshots. There is one capability note for prompt-fidelity users: both 2.0 tiers accept negative_prompt, and the 3.0 series supports it as a standard optional parameter — check your exact model’s reference before relying on it. And if you are evaluating Qwen against non-Qwen alternatives side by side, the unified AI API guide explains how to run multi-vendor comparisons on one key.

Calling Qwen Image 2.0 through Modellix: the honest answer

We checked our own catalog on September 6, 2026, and we are not going to dress it up: Modellix does not serve Qwen Image 2.0 anymore. The model pages for qwen-image-2.0 and qwen-image-2.0-pro now render “Model Not Found” — there is nothing live behind that URL to price, and the honest current option on our side is the Qwen Image 3.0 family. This article deliberately shows no Modellix price for 2.0.

What Modellix does serve, as of today, is the current generation: qwen-image-3.0 at a displayed $0.0270/image and qwen-image-3.0-pro at $0.0675/image, plus qwen-image-3.0-edit and -pro-edit for image-to-image work — all behind one API key with the same https://api.modellix.ai/api/v1/<provider>/<model> convention — media routes take the documented /async task form — documented in the Modellix API docs. A generation call looks like this (async — first call returns a task handle, then you poll or set a webhook):

curl --request POST \
  --url https://api.modellix.ai/api/v1/alibaba/qwen-image-3.0-pro/async \
  --header 'Authorization: Bearer ***' \
  --header 'Content-Type: application/json' \
  --data '{
    "prompt": "A developer-conference poster titled Build With Qwen Image 2.0, bold white sans-serif on an amber and cyan gradient",
    "size": "2048*2048",
    "prompt_extend": true
  }'

Qwen Image API Reference

See the full request and polling pattern for Qwen image models on Modellix, including 3.0, 3.0-pro, and the edit variants.

View Docs

Why would an integrator who came for 2.0 accept 3.0 on an aggregator instead of calling Alibaba direct? The honest reasons: one key for Qwen and the rest of Modellix’s 210+ image and video models, no Chinese mainland account requirement, per-call cost logs, and pay-as-you-go billing. The honest boundary: for a single high-volume model, direct Model Studio billing can win on unit price — at 1K resolution, official 3.0-pro is $0.043/image including input versus Modellix’s single displayed $0.0675, while at 2K official 3.0 is $0.033 including input versus Modellix’s $0.0270 — the cheaper route genuinely depends on the resolution you generate at. The aggregator’s value is operational, and we say so plainly rather than claiming a universal discount.

Run the Current Qwen Image Family on One Key

Log in to Modellix to call Qwen Image 3.0 and 210+ other image and video models with one API key.

Login

Frequently Asked Questions

Is Qwen Image 2.0 deprecated or still supported?

Still supported as of September 6, 2026. Alibaba Cloud Model Studio continues to list qwen-image-2.0 and qwen-image-2.0-pro on its pricing and API pages without a legacy label. The honest caveat: the last 2.0 snapshot is dated June 2026 — since Qwen Image 3.0 launched in July 2026, no new 2.0 releases have shipped, so treat it as frozen-but-served, not deprecated.

What is the Qwen Image 2.0 API model ID?

qwen-image-2.0 (accelerated tier) and qwen-image-2.0-pro (Pro tier). Dated snapshot IDs (qwen-image-2.0-2026-03-03, qwen-image-2.0-pro-2026-03-03/-2026-04-22/-2026-06-22) pin a specific release.

How much does the Qwen Image 2.0 API cost?

$0.035 per successfully generated image for qwen-image-2.0 and $0.075 for qwen-image-2.0-pro, Singapore region, as of September 6, 2026. Only successful images are billed.

Is there a free Qwen Image 2.0 API?

Model Studio grants 100 free images for both 2.0 tiers, Singapore region only, valid 90 days from account activation.

What is the Qwen Image 2.0 API base URL?

https://{WorkspaceId}.ap-southeast-1.maas.aliyuncs.com/api/v1/services/aigc/multimodal-generation/generation (Singapore) or the cn-beijing equivalent. Replace {WorkspaceId} with your workspace ID and use a matching region key.

What is the difference between qwen-image-2.0 and qwen-image-2.0-pro?

qwen-image-2.0 balances quality and speed at $0.035/image; 2.0-pro adds stronger text rendering, realism, and semantic adherence at $0.075/image. Both support generation and editing, 1–6 images per call, and free-form sizes up to 2048×2048.

Can I edit images with the Qwen Image 2.0 API?

Yes — generation and editing are unified in the 2.0 model. Editing uses the same multimodal-generation endpoint with input images plus an instruction, at the same per-image prices.

Should I use Qwen Image 2.0 or 3.0 in 2026?

Choose 2.0 when your pipeline is already integrated and stable, you want a flat per-image price with no input charge, or you do not need dense multi-panel layouts. Choose 3.0 for new integrations, infographics, small multilingual text, or long structured prompts (up to ~4,500 tokens).

Can I call Qwen Image 2.0 through Modellix?

No — as of September 6, 2026, Qwen Image 2.0 has been removed from Modellix’s catalog (its model page renders “Model Not Found”). Modellix serves the current Qwen Image 3.0 family, including 3.0, 3.0-pro, and the edit variants.

---

Qwen Image model status, pricing, and availability reflect public information as of September 6, 2026 and change frequently — Alibaba has already moved models between generations within a single quarter. Re-validate against the official Model Studio pricing and model pages before committing budget. This article was written by Modellix, an API aggregator with a commercial interest in the Qwen Image 3.0 family it routes; the Qwen Image 2.0 integration material above is vendor-neutral. Access 210+ image and video models through one API key at modellix.ai.