MemTensor releases Metis models that write memory inside the backbone
The Qwen3.5-based research preview includes 4B, 9B and 27B checkpoints, while withholding training data and restricting the code to noncommercial use.
By Ryan Merket · Published
Primary source: X
Why it matters
Persistent state is becoming a core constraint for agents that operate across days and applications. Metis tests whether some of that state can move from vector databases and prompts into the model's computation itself.
MemTensor, the Shanghai AI developer founded by former Alibaba data executive Xiong Feiyu, released a family of models designed to maintain memory inside the model backbone instead of repeatedly retrieving old text from an external database.…
When you purchase through links in our articles, we may earn a small commission. This doesn’t affect our editorial independence.