Alibaba Drops Qwen3.8 Flash Next
The model uses a new design with 125 billion parameters in total. Even with that high count, it only uses 6 billion parameters for each word it processes. This makes it very fast and much cheaper to train than previous versions.
Technically the system relies on new components like the Gated DeltaNet and an N-gram embedding table. These additions help the model handle complex tasks while keeping computing costs low. It is a major shift in how they build their architecture.
Keep in mind that running this model yourself requires a lot of power. The file size is over 170 gigabytes. You will need high-end hardware and plenty of memory to get it running on your own servers.
Comments (0)
No comments yet. Be the first!
More AI news
NewsGoogle AI Changes Its Search Advice After Bias Complaints
Google updated its search tool after it incorrectly told users to call emergency services based on a person's nationality.
NewsWhy AI Is Still Failing at Simple Tasks
Researchers gave an AI five thousand dollars to grow, but it could not even open a bank account.
NewsEnovis to Buy eCential Robotics
Enovis is expanding its surgical tech business by purchasing French company eCential Robotics.