Qwen vs ChatGPT: What Qwen 3.8 and GPT-6 Astra Do, What They Cost, and Who Each Is For

Pick ChatGPT if you want one ready-made app for everyday work: the free plan covers chat, image creation, voice and some deep research, and Plus (the $20-a-month plan) adds GPT-6 Astra. Pick Qwen if you want Alibaba's free chat assistant or a model you can download and run yourself, since Qwen3.8-27B is published under the Apache 2.0 license, which allows commercial use. Pat Simmons hasn't tested Qwen against ChatGPT, so everything on this page comes from OpenAI's, Alibaba's and Hugging Face's own pages.

Last checked November 9, 2026: ChatGPT's plans, prices and models on OpenAI's pricing page (US prices), the newest Qwen releases on Qwen's Hugging Face page (Qwen 3.8 has been out since August 2026), and Qwen's model cards and Qwen Cloud prices.

Qwen vs ChatGPT: our verdict, by job

ChatGPT suits people who want the newest OpenAI models and a long list of built-in tools in one subscription. Qwen suits people who want a free assistant, a model they can host themselves, or a low per-use price for building their own app.

Everyday chat, writing and research in one app: ChatGPT. Its plans bundle file uploads, voice, image creation, deep research, memory and projects, and Plus adds the newest models (GPT-6.1 Sol, GPT-6 Astra and GPT-6 Sol) plus Sites, according to OpenAI's plan comparison. Qwen's own chat box offers Search, Web Dev, Artifacts, Deep Research, Image Generation and Video Generation, so it covers a lot of the same ground for free.

Paying nothing: both work. ChatGPT Free costs $0 and gives unlimited text chats with GPT-6 (OpenAI adds that "abuse guardrails" apply), with limited uploads, image creation, voice chats and deep research. Alibaba describes its Qwen assistant as "free to use, open to all."

Running the AI on your own computer or company servers: Qwen. Alibaba publishes the model files on Hugging Face, and Qwen3.8-27B's license lets you use it at work and in products you sell. The section on running Qwen yourself below covers what the model card does and doesn't tell you.

Reading very long documents: close on paper. Qwen3.8-27B's model card lists a context window (how much text the model can hold at once) of 262,144 tokens, extendable to about 1 million. ChatGPT Plus gives its reasoning models a 256K window, which OpenAI puts at about 320 pages of input, and Pro raises that to 400K.

Building your own app on a cheap model: Qwen Cloud, Alibaba's paid service for developers, lists a per-use price for Qwen3.8-27B that's covered in the cost section. OpenAI's developer prices weren't checked for this page, so there's no per-token comparison here.

Is Qwen better than ChatGPT?

Nobody has published a fair, same-test comparison of Qwen 3.8 and ChatGPT's GPT-6 models, and AI for Mortals hasn't run one. None of the Qwen 3.8 model cards includes a GPT-6 model in its scores.

The closest thing is the table on Alibaba's Qwen3.8-2.4T-A95B model card, Qwen's largest open model, which compares its hosted version, Qwen3.8-Max, with GPT-5.6 Sol, an older OpenAI model that ChatGPT Plus still lists. On SWE-bench Pro, a test of fixing real software bugs, Alibaba reports 67.7 for Qwen3.8-Max and 64.6 for GPT-5.6 Sol, after running every model on its own corrected version of the test.

The same table gives GPT-5.6 Sol the edge on GPQA Diamond, a set of graduate-level science questions, at 94.1 to Qwen3.8-Max's 92.6. These are Alibaba's own numbers, with Alibaba choosing the tests and settings, and for some rows the footnotes say it copied rivals' scores from their published results.

For ChatGPT's newest model in real use, Pat has put GPT-6 Astra through head-to-head builds: see how GPT-6 Astra did against Claude and Gemini on the same builds, and Grok vs ChatGPT. Those posts show how GPT-6 Astra handled each build, and neither one includes Qwen.

What size Qwen models are there?

Qwen 3.8 comes in three open sizes: 27B, Flash-Next and 2.4T-A95B. Qwen 3.6 comes in 27B and 35B-A3B, and Qwen 3.5's card shows 27B, 35B-A3B and 122B-A10B models in its own score table.

The "B" stands for billions of parameters, the numbers a model learns in training. A name like 35B-A3B means the model holds 35 billion parameters in total but activates only 3 billion of them at a time, a design Alibaba says keeps answers fast and cheap to produce.

Alibaba calls Flash-Next an "experimental preview" of the design that will underpin Qwen 4. Its hosted service runs official versions of two of these: Qwen3.8-Max, built on the 2.4T model with image input added, and Qwen3.8-Flash, built on Flash-Next. On Alibaba Cloud Model Studio, Alibaba describes qwen3.8-max as for "Complex reasoning and coding," qwen3.7-plus as "Balanced performance, speed, and cost" and qwen3.8-flash as "Fast and cost-effective."

Why is Qwen good?

Qwen pairs high self-reported scores with licenses that let almost anyone download and use it for free, including at work. Every score below is Alibaba's own.

Strong scores for a compact model: on the Qwen3.8-27B model card, Alibaba reports 89.2 on GPQA Diamond and 90.3 on LiveCodeBench v6, a set of coding puzzles. That table compares it with earlier Qwen models, Muse Glimmer-30B and Opus 4.6 Max, which scores 91.3 on GPQA Diamond.

Licenses that allow business use: Qwen3.8-27B and the Qwen 3.6 and 3.5 models use Apache 2.0, a standard open license that allows commercial use. The two bigger Qwen 3.8 models have their own licenses. The Flash-Next license asks any company that sells access to AI models, or sells an AI coding or office assistant, to get a separate license before commercial use. The 2.4T model's license sets that bar at more than US$50 million in revenue over 12 months, and both ask products with over 100 million monthly users or US$20 million in monthly revenue to show the model's name.

Images and long video: Qwen3.8-27B reads images and video, and Alibaba says that runs "from STEM diagrams and documents to hour-scale videos." For audio, Model Studio points to a separate model, qwen3.8-omni-flash.

Many languages: the Qwen 3.5 model card lists support for 201 languages and dialects.

Thinking you can switch off: Qwen3.8-27B works through a problem step by step before answering by default, and developers can turn that off for a faster direct reply.

For how Qwen compares with Google's open model, see Gemma 4 vs Qwen 3.5, 3.6 and 3.8, side by side.

How much do Qwen and ChatGPT cost?

ChatGPT costs $0 on Free, $8 a month on Go, $20 a month on Plus and from $100 a month on Pro. Qwen's chat assistant and its downloadable models are free, and developers who use Qwen through Alibaba's cloud pay per use.

ChatGPT prices are as shown on OpenAI's US pricing page on November 9, 2026. Tokens are the chunks of text an AI model reads and writes, and developers pay for both what they send and what the model sends back.

The Hugging Face card for Qwen3.8-27B still calls the hosted version "coming soon," but the Qwen Cloud page above already lists it with prices. For more paid and free options, see other ChatGPT alternatives and what they cost.

Can you run Qwen on your own computer?

Yes. Alibaba publishes Qwen's model files on Hugging Face, and the Qwen3.8-27B page there lists 1,729 quantized versions, which are compressed copies that need less memory, with buttons to open them in LM Studio, Jan and llama.cpp.

The model card doesn't say how much memory or what graphics card you need. It points developers to server software (SGLang, vLLM and TokenSpeed) for running the model at scale, and Alibaba recommends its own cloud for anyone who wants it hosted.

The Qwen3.6-35B-A3B card lists the same 262,144-token context window and the same Apache 2.0 license, and it activates only 3 billion of its parameters at a time.

Questions people ask

Is Qwen free?

Yes. Alibaba's Qwen homepage describes the Qwen assistant as "free to use, open to all," and the Qwen 3.5, 3.6 and 3.8 model files are free to download. Only Alibaba's cloud service for developers charges per use.

Is ChatGPT free?

Yes. ChatGPT Free costs $0 and runs on GPT-6 Luna, with limits on uploads, image creation, voice and deep research. GPT-6 Astra needs Plus or Pro.

Qwen 3.8 vs GPT-6: which scores higher?

Neither company has published a comparison of the two. Alibaba's Qwen 3.8 tables stop at GPT-5.6 Sol, so any Qwen 3.8 vs GPT-6 call today would mix numbers from different tables run in different ways.

Who makes Qwen?

Alibaba. Qwen Cloud lists Alibaba as the model author, and the models are sold through Alibaba Cloud Model Studio.

Can Qwen understand video?

Yes. Qwen3.8-27B takes text, images and video, including videos about an hour long, and answers in text.

Can ChatGPT and Qwen read whole books?

Qwen3.8-27B holds 262,144 tokens natively. On ChatGPT, OpenAI puts the most you can paste into a reasoning model at about 320 pages on Plus and about 680 pages on Pro, and about 12 pages for the faster Instant model on Free.

Where can I follow new releases?

The AI for Mortals tool pages track ChatGPT's latest features and Qwen updates and new releases.

How this comparison was put together

No test was run for this page. ChatGPT's plans, prices, models and limits come from OpenAI's pricing page, read on November 9, 2026. Qwen's sizes, dates, licenses, context windows and scores come from Alibaba's model cards on Hugging Face, and its hosted prices and model descriptions from Qwen Cloud and Alibaba Cloud Model Studio, all checked the same day.

Every benchmark score here is self-reported by Alibaba, which picks the tests, the settings and the rivals in its own tables. Pat Simmons's own GPT-6 Astra tests are linked above as a guide to ChatGPT, and they are not evidence about Qwen.

Sources

• OpenAI: ChatGPT plans and pricing (checked November 9, 2026)
• Qwen: Qwen3.8-27B model card (checked November 9, 2026)
• Qwen: Qwen3.8-2.4T-A95B model card and license (checked November 9, 2026)
• Qwen: Qwen3.8-Flash-Next model card and license (checked November 9, 2026)
• Qwen: Qwen3.6-35B-A3B model card (checked November 9, 2026)
• Qwen: Qwen3.6-27B model card (checked November 9, 2026)
• Qwen: Qwen3.5-27B model card (checked November 9, 2026)
• Qwen on Hugging Face: model list (checked November 9, 2026)
• Qwen Cloud: Qwen3.8-27B pricing (checked November 9, 2026)
• Alibaba Cloud Model Studio: recommended models (checked November 9, 2026)
• Qwen: homepage (checked November 9, 2026)
• Qwen Chat (checked November 9, 2026)