Qwen-Plus: Alibaba Cloud's Mid-Tier API Model

Qwen-Plus is Alibaba Cloud's mid-tier API model — the balanced option between the budget-friendly Qwen-Flash and the frontier-grade Qwen-Max. It offers strong reasoning, 1M-token context, and thinking mode at a fraction of the cost of competing frontier APIs. As of February 2026, the Plus tier includes Qwen3.5-Plus — powered by the newly released Qwen 3.5 open-source model — alongside the stable Qwen3-based qwen-plus alias.

What Is Qwen-Plus?

Qwen-Plus is not a specific open-source model — it's an API tier alias on Alibaba Cloud's Model Studio (DashScope). Think of it like OpenAI's "GPT-4o" vs "GPT-4o-mini" — different price/performance tiers pointing to different underlying models. The qwen-plus alias always points to a stable, production-ready snapshot that Alibaba updates periodically.

The key advantage of Qwen-Plus over open-source self-hosting: you get 1M-token context, thinking mode, and production-grade infrastructure without managing GPU servers. The key advantage over Qwen-Max: it's 3× cheaper on input and significantly faster for most tasks.

API Tiers: Flash vs Plus vs Max

Alibaba offers three main API tiers. Understanding the differences helps you pick the right one:

Note: Qwen-Turbo is deprecated. Alibaba recommends migrating to Qwen-Flash for equivalent low-cost use cases.

Qwen-Plus Specifications

Pricing

Qwen-Plus (stable, Qwen3-based)

Qwen3.5-Plus (new, Qwen3.5-based)

• Batch processing: 50% discount for asynchronous batch jobs.
• Context caching: Implicit cache hits cost 20% of standard input; explicit cache hits cost 10%.
• Free tier: 1 million tokens for new international accounts (valid 90 days).

How does this compare? Qwen-Plus at $0.40/$1.20 per 1M tokens is roughly 5–10× cheaper than GPT-4o and 3× cheaper than Qwen-Max on input costs. For most production workloads that don't require frontier-level reasoning, Plus offers the best cost-to-quality ratio in Alibaba's lineup.

Qwen3.5-Plus — The New Option

Released on February 15, 2026, Qwen3.5-Plus is powered by the Qwen 3.5 open-source model (397B parameters, 17B active). Key differences from the standard qwen-plus:

• Multimodal input: Accepts text, images, and video (standard qwen-plus is text-only).
• 201 languages (vs ~100 in standard qwen-plus).
• Stronger benchmarks: Leads on IFBench, MathVision, OmniDocBench, and all agentic benchmarks.
• No separate thinking surcharge: Same price regardless of thinking mode (standard qwen-plus charges 3.3× more for thinking output).
• Slightly higher output cost: $2.40/M vs $1.20/M for standard mode — but no thinking surcharge makes it cheaper for reasoning tasks.

Which should you use? If you need multimodal capabilities, agentic features, or run heavy reasoning tasks, qwen3.5-plus is the better choice. If you have a stable production pipeline that only needs text processing and you want the lowest cost, stick with qwen-plus until Alibaba updates the alias.

Model IDs

• qwen-plus / qwen-plus-latest — Stable alias (currently points to Qwen3-based snapshot)
• qwen3.5-plus-2026-02-15 — Qwen 3.5 specific snapshot

API Quick Start

The API is OpenAI-compatible. You can use the OpenAI Python SDK with a different base URL:

Basic Text Request

from openai import OpenAI

client = OpenAI(
    api_key="YOUR_DASHSCOPE_API_KEY",
    base_url="https://dashscope-intl.aliyuncs.com/compatible-mode/v1"
)

response = client.chat.completions.create(
    model="qwen-plus",  # or "qwen3.5-plus-2026-02-15"
    messages=[
        {"role": "user", "content": "Summarize the key risks in this contract..."}
    ]
)

print(response.choices[0].message.content)

With Thinking Mode

response = client.chat.completions.create(
    model="qwen-plus",
    messages=[
        {"role": "user", "content": "Solve this optimization problem step by step..."}
    ],
    extra_body={"enable_thinking": True}
)

Multimodal (Qwen3.5-Plus only)

response = client.chat.completions.create(
    model="qwen3.5-plus-2026-02-15",
    messages=[{
        "role": "user",
        "content": [
            {"type": "image_url", "image_url": {"url": "https://example.com/chart.png"}},
            {"type": "text", "text": "What does this chart show?"}
        ]
    }]
)

Version History

The qwen-plus alias has pointed to different underlying models over time as Alibaba upgrades the backend:

When to Use Plus vs Max vs Flash

FAQ

Is Qwen-Plus the same as Qwen3.5-Plus?

Not exactly. qwen-plus is the stable alias that Alibaba updates periodically (currently Qwen3-based). qwen3.5-plus is a specific named model based on the Qwen 3.5 architecture. They coexist — you can choose either via the API model ID.

Can I run Qwen-Plus locally?

No — Qwen-Plus is API-only. However, the underlying models are open-source. The Qwen3.5-Plus API runs the same Qwen 3.5 model (397B MoE) that's available on HuggingFace under Apache 2.0. You can self-host it if you have the hardware (~256GB+ RAM). See our Run Locally guide.

How does Qwen-Plus compare to GPT-4o?

Qwen-Plus offers comparable quality for most enterprise tasks at 5–10× lower cost. The Qwen3.5-Plus variant adds multimodal capabilities and stronger benchmark scores than GPT-4o on instruction following and document understanding. For tasks requiring frontier reasoning (math, science), Qwen-Max is the better comparison to GPT-5.

What happened to Qwen-Turbo?

Deprecated. Alibaba recommends migrating to Qwen-Flash, which serves the same low-cost, high-speed niche at $0.05/M input tokens.