MemOS

Headlines · 3

1. Oct 20
   MemT 1.0 Opens the Source for Memory-Native AI Models

   MemTensor released and open-sourced MemT 1.0, a family of memory-native foundation models comprising MemT-1.0-Nano and MemT-1.0-Flash. The architecture uses Gated DeltaNet-2 and Mixture-of-Storage to maintain updating internal memory states that participate in generation, with author-reported gains in long-history inference efficiency, storage reduction, and agentic task success. External text-memory systems still retain an advantage for precise recall from long histories, according to the reports.

   Hugging Face · Papers · 🔥 4
2. Oct 19
   Tencent Hunyuan releases GGUF quantizations of Hy-Memory-30B-A3B for llama.cpp

   Tencent Hunyuan published a GGUF repository for Hy-Memory-30B-A3B, a roughly 30B-parameter long-term-memory writer model with about 3B active parameters, aimed at local inference through llama.cpp. Three quantization levels are offered: Q6_K imatrix at 24.71 GB, Q8_0 at 31.99 GB and Q4_K_M imatrix at 18.24 GB, all under Apache-2.0. The model card reports LoCoMo Judge scores of 58.78% (154/262) for Q6_K, 57.63% for Q8_0 and 56.49% for Q4_K_M on a fixed 20% LoCoMo split using the Mem0 2.0.19 harness, with gpt-5.6-sol as answer and judge model; these figures are self-reported by the publisher and not independently replicated.

   腾讯混元 · Hugging Face · 🔥 0
3. Oct 16
   Tencent Hunyuan open-sources Hy Memory 2.0, Hy-Memory-30B-A3B and MemHub

   Tencent Hunyuan announced Hy Memory 2.0 on October 16, open-sourcing three layers: the dedicated memory model Hy-Memory-30B-A3B, the memory framework SDK and the local MemHub app. Tencent says the framework handles chat, work and code memory and ranks first against open frameworks including Mem0, MemOS, EverOS and OpenViking on four benchmarks; it reports that Hy-Memory-30B-A3B, with about 3B active parameters, scored a 55.49 macro average across four memory harnesses versus 54.62 for Claude Opus 5. MemHub ships as a CLI and Mac app that can connect to roughly 10 tools such as WorkBuddy, Cursor, Codex and Qwen Code, while the model weights are published on Hugging Face under Apache 2.0, followed on October 18 by an FP8 version optimized for vLLM deployment. All performance figures are Tencent's own claims and have not been independently verified.

   腾讯混元 · Hugging Face · 🔥 0

Experience and discussion from the community

Share my MemOS experienceAsk about MemOS