Memtensor Releases Metis: Memory Foundation Model Without RAG
AdinaYakup · x · 2026-07-31
Memtensor Research Group released Metis, a memory foundation model claimed to be the first LLM that internalizes memory directly into its backbone.
Key features include:
- Available in 4B/9B/27B parameter weights, with code and paper open-sourced.
- Eliminates the need for external RAG, as memory is stored inside the model.
- Completes full memory read/write in a single forward pass.
- Weights are frozen during inference, allowing deployment like a standard LLM.
Related event: Metis Memory LLM Released as Open Source(2 posts)→
More from Models
- HuggingFace Repelled Proprietary Model Attack Using Open Source Model — huggingface · 2026-07-31
- Testing Google Gemini 3.5 Flash: 8x Speed Increase and General Improvements — Zergylord · 2026-07-31
- Meituan Open-Sources LongCat-Flash-Lite-Sparse: 256k Context on a 24GB GPU — Gohab2001 · 2026-07-31
- Merge Launches Fusion API: Parallel Multi-Model Panel with Judge Claims to Beat Frontier Models — shensi · 2026-07-31
- Developer Calls for Dynamic Presets in Reasoning Models to Switch Inference Effort — intellectronica · 2026-07-31
- OpenAI and Anthropic Lack Time to Build Good Security Sandboxes, Researcher Warns — peterwildeford · 2026-07-31