Qwen vs ChatGPT: GPT scores higher for a seventh of the cost per task, and Qwen's free app is for non-commercial use only

Is Qwen better than ChatGPT?

No, Qwen isn't better: GPT-6.1 Sol beats Qwen3.8 Max on all three things that Artificial Analysis measures. It puts each model through the same ten tests, from Terminal-Bench 4.0 to Humanity's Last Exam, then prices one task at list rates and times it. Sol, the newest model on ChatGPT Plus, ran at max effort and took 570 seconds a task. Qwen3.8 Max, the best that Alibaba has, took 2,490 seconds, which is about 41 minutes. So OpenAI gives you seven more points for about a seventh of the price and under a quarter of the wait.

At the cheap end of each family, it's much closer. Qwen3.8 Flash Next scores 40 for $0.37 a task. GPT-6 Luna, which is the model that OpenAI's pricing page lists on the free ChatGPT plan, scores 38 for $0.07. That's two points to Qwen at five times the cost.

All of the scores and task costs come from the Artificial Analysis leaderboard (Intelligence Index v4.3.2), and we read them on 12 October 2026. The OpenAI prices are from its API pricing page. The Qwen3.8 Max and Qwen3.7 Plus prices are from the Model Studio price list for Alibaba's Singapore region, which it updated on 12 October 2026. The Flash Next price is the one Artificial Analysis lists, and it matches qwen3.8-flash on Alibaba's list. When the weights of a model are open, any host can set its own price for it.

Why does Qwen cost more per task when its tokens are cheaper?

Qwen3.8 Max costs more per task because it thinks for a lot longer. Per token, it's the cheaper of the two: Alibaba charges $2 for each million tokens you send to it and $6 for each million that it writes, and OpenAI charges $2 and $10 for GPT-6.1 Sol. But Artificial Analysis counted 108,000 output tokens per task for Qwen3.8 Max, and 71,000 of them were spent on reasoning. Sol wrote 38,000.

Cached input makes the gap wider. OpenAI bills a cached token on Sol at $0.10 per million. Artificial Analysis lists $0.25 for one on Qwen3.8 Max, and Alibaba's own cache page says only that this model doesn't get the usual cache discount. An agent task reads the same context again and again, so that line adds up. It's the same at the cheap end, where Qwen3.8 Flash Next wrote 108,000 tokens for each task and GPT-6 Luna wrote 50,000.

Alibaba gives each new account in Singapore 1M free tokens on each of qwen3.8-max, qwen3.8-flash and qwen3.7-plus, and they last for 90 days. It also sets its prices by region, and its China (Beijing) list is a separate one. If you're building on either API, price the job, not the token.

Is Qwen better than ChatGPT for coding?

GPT-6.1 Sol is the better one for coding. On Terminal-Bench 4.0, which tests an agent as it works in a real terminal, it scores 56% and Qwen3.8 Max scores 39%. The two of them are close on SciCode, at 54% to 52%. Sol is also ahead on AutomationBench-AA, by 65% to 56%.

The cheap models split it. Qwen3.8 Flash Next scores 25% on Terminal-Bench to the 13% of GPT-6 Luna, so Qwen is the one for a terminal agent on a budget. Luna wins on SciCode, by 55% to 51%. It's also quicker, at 138 tokens a second to the 51 of Qwen, and it finished each task in 360 seconds where Qwen took 1,860. You'll feel that in a coding agent that loops for an hour.

Is Qwen or ChatGPT better in other languages?

GPT beats Qwen in other languages, and by a wide margin. On the Whizi Multilingual LLM Bench, GPT-6 Sol passed 95 of the 150 questions (63.3%) and Qwen3.8 Max passed 46 of them (30.7%). We wrote all of the questions from scratch, five in each of 30 languages, and we graded every answer against a key with a source. Each of the models ran once at low reasoning, and the run was over on 2 October 2026.

The gap between Sol and Qwen3.8 Max is 33 points, which is far wider than the range of about 7 points on either of them, so it isn't noise. Even GPT-6 Luna, the cheap model from OpenAI, scored 14 points above the flagship from Qwen. Qwen didn't win in Chinese either. There are ten Chinese questions in the bench, five in simplified and five in traditional script, and Qwen3.8 Max passed six of them to the eight that Sol passed. Ten is too few to rank anyone on Chinese alone. It's enough to say that Qwen showed no home advantage.

Is Qwen Chat free, and what does ChatGPT cost?

Qwen Chat is free. It now goes by Qwen Studio (Qwenstudio in its terms), and it still runs at chat.qwen.ai. As of October 2026 Qwen publishes no price list for it and no fixed message limit. Its terms of service already describe paid tiers, though, each with "a usage allowance for each subscription period", billed through Apple, Google or Stripe.

Read those terms before you use it for your job. They say: "The Services are for private and non-commercial use only." They also say you must be at least 18. The OpenAI terms of use, which have been in effect since 1 January 2026, don't have a limit like that. They set the age at 13, and they give you the rights to what ChatGPT writes for you.

ChatGPT Free gives you unlimited everyday text chats, "subject to abuse-prevention safeguards", in the words of the OpenAI free tier FAQ. It may show you ads in some countries. Go is $8 a month in the US, as OpenAI said when it launched Go there, and it raises the caps on images, files and data analysis. Plus is $20 a month, and that's the plan with GPT-6.1 Sol. Pro starts at $100. ChatGPT free vs paid lists what each tier unlocks.

Where do Qwen and ChatGPT keep your chats?

Qwen keeps them in Singapore, and ChatGPT keeps them in the United States. The Qwen privacy policy names NTH Power Global Tech Singapore as the company that runs the app and the qwen.ai website. It says that Qwen collects, processes and stores your personal data in Singapore. OpenAI's privacy policy describes "processing and storing your Personal Data in our facilities and servers in the United States". Both policies add that your data may also be processed in other countries.

Both of them use what you type to make their models better, but in different ways. Qwen says that it may train on content and feedback once it has been de-identified, and that it won't use your raw input to train a model directly. It gives you no switch to turn that off, only a general right to object to how it handles your data. OpenAI may train on your content too, and its data controls FAQ shows you where the switch is: Settings, then Data controls, then turn off "Improve the model for everyone". A Temporary Chat is kept for up to 30 days.

Can you run Qwen or GPT on your own computer?

You can run both on your own computer, and this is where Qwen wins clearly. Qwen3.8 27B is out under Apache 2.0 on Hugging Face, with 262,144 tokens of context that you can stretch to 1M. It scores 34 on the Intelligence Index. The best open model that OpenAI has, gpt-oss-120b, is under Apache 2.0 too, but it's from August 2025 and it scores 12.

The bigger open models from Alibaba come with strings. Qwen3.8 Flash Next is under the Qwen Community License 1.0, and the 2.4T A95B model has a Qwen3.8-Max License of its own. Both of them ask some of the businesses that sell an API or a coding tool to sign a separate deal. OpenAI keeps all of its GPT-6 models closed. If your data can't leave your own machine, Qwen is the family for you, and the open source LLMs guide covers how to run one.

Should you pick Qwen or ChatGPT?

Pick ChatGPT if you want the best answers for each dollar, if it's for your job, or if you chat in many languages. Pick Qwen if you want it free, if you want the weights, or if your data has to stay out of the US.

Whizi has both of these families, and on Pro the GPT model is the one to pick. Pro is $29.99 a month with 2,500 credits. It has GPT-6.1 Sol at 10 credits a message and Qwen3.7 Plus at 1, and its side by side view sends one prompt to both of them. But Qwen3.7 Plus scores 25. Qwen3.8 Max is on Powerhouse at $49.99, and it costs 6 credits a message there. The GPT and Qwen docs list all of the models. DeepSeek vs ChatGPT runs the same match against China's other big lab.