Alibaba Qwen3.8-Max free release is planned, not yet complete
Alibaba announced Qwen3.8-Max and plans to publish it for download, but its performance claims have not yet been independently tested.
By Theo Nakamura · Staff Writer
· 3 min read
The Alibaba Qwen3.8-Max free release had not happened as of Aug. 4, despite headlines suggesting the model was already available to download. Alibaba announced its new flagship AI model on Aug. 3 and said it would publish the model the following week, according to Forbes and Aivy.
That timing matters for investors watching Alibaba’s AI efforts. The company has introduced a model it says can compete with leading U.S. systems, but the planned downloadable version, independent testing and the eventual licensing terms were still pending in the available reporting.
When will Alibaba Qwen3.8-Max be free to download?
Alibaba had promised to publish Qwen3.8-Max free in the following week, Aivy reported. As of Aug. 4, users could access the model through Alibaba Cloud’s service, a work assistant and OpenRouter, but could not download it.
An open model allows developers to download, run and modify it, Forbes reported. That differs from proprietary systems such as ChatGPT and Claude, whose underlying systems are private. Alibaba had not yet published the downloadable model weights, the files that enable others to run the model themselves, according to Aivy.
Hosted access was not free. Aivy reported listed rates of $2 per million input tokens and $6 per million output tokens. Tokens are chunks of text a model processes, and the pricing is listed separately for material sent to the model and material it returns.
What does Alibaba claim Qwen3.8-Max can do?
Alibaba called Qwen3.8-Max its most capable AI model to date, according to Forbes. The company said the model has 2.4 trillion parameters, the internal values that help determine how a model processes information, placing it among the largest open AI models described in the report.
Alibaba said Qwen3.8-Max is designed for coding, workplace tasks, research and long-horizon execution, meaning work that requires a series of steps over an extended period. Euronews reported that Alibaba said one test had the model operate autonomously for about 125 hours while rebuilding and improving a mathematics research experiment. Those are Alibaba’s claims, rather than independently established results.
How close is Qwen3.8-Max to Claude and ChatGPT?
Alibaba published results that it said showed Qwen3.8-Max outperforming Anthropic’s Fable 5 in selected areas and reaching results comparable to or higher than OpenAI’s GPT5.6-Sol on some measures, Forbes reported. Those comparisons should be read as company-reported benchmarks, not confirmed parity.
Aivy reported that no independent benchmark provider had published a score or dedicated page for Qwen3.8-Max as of Aug. 4. It also reported that Alibaba used its own testing setup for Qwen3.8-Max, relied on published leaderboard results for rivals and included tests it created itself.
The disclosed figures were mixed. In one programming test cited by Aivy, Alibaba’s model scored 67.7, compared with 80.0 for Claude Fable 5. Independent evaluations will be needed to judge how the model performs across coding and other real-world uses.
This story draws on original reporting from Decrypt.
