MemOS launches MemT 1.0: memory-native LLMs that survive context clears
lxfater · x · 2026-10-21
MemOS team released MemT 1.0, a memory-native model line (Flash 288B, Nano 36B) with built-in memory read/write modules that store history as numerical states instead of raw chat logs.
- At 128K history length, Nano runs 1.86x faster end-to-end than full-context, with persistent state storage 1/66 of the original.
- Injecting experience distilled from past coding tasks lifts SWE-bench Verified success from 66.07% to 69.33%.
- GitHub demo shows saving memory, clearing context, restoring it, and still answering earlier questions.
The takeaway: persisted context and corrections could mean less repetition and fewer detours in future conversations.
Related event: MemTensor Open-Sources Memory-Native Model MemT-1.0(10 posts)→
More from Models
- Testing Jev, an RLCD decision model, to build topical maps of Anthropic's 547-page blog — cneuralnetwork · 2026-10-21
- Which cheap LLM API is reliable for function calling and structured JSON? — Engeneer_Matt · 2026-10-21
- Assistant market share shakeup: ChatGPT falls 71.8% to 56.4% as Gemini and Claude surge — Linahuaa · 2026-10-21
- Musk: Grok Bot is 'the fastest-growing product I've ever seen in my life' — elonmusk · 2026-10-21
- NVIDIA releases seven open model families at GTC Berlin, from Nemotron 3.5 to Cosmos 3.5 and BioNeMo — NVIDIAAI · 2026-10-21
- NVIDIA Expands Open Model Family: Nemotron 3.5, Cosmos 3.5, Kumo, BioNeMo and Ising — NVIDIA Blog · 2026-10-21