Is MemT-1.0 state of the art?Pending benchmarks.Announced Oct 20, 2026; no independent results yet.

The details so far

Lab

MemTensor

Announced

Oct 20, 2026

API

Not yet

Open weights

Published

Parameters

Flash ~288B total / ~17B active; Nano ~35B total / ~3B active

Lab

MemTensor (记忆张量), China

Variants

MemT-1.0-Flash (built on DeepSeek-V4-Flash) and MemT-1.0-Nano (built on Qwen3.6-35B-A3B)

Memory

Dynamic local memory plus trainable hyper memory; locally persistent memory sessions

Weights license

Flash: MIT. Nano: Apache-2.0 (Hugging Face cards). Inherited backbone licenses still apply

Code license

GitHub repo MemTensor/MemT-1.0: PolyForm Noncommercial 1.0.0 (repo LICENSE)

API

Not mentioned

Tags marked unsure are not stated on the lab's own pages.

What would let us call it

Verdicts use only independent boards we already track. Two consecutive snapshots are required before any crown.

• Publicly announced with a source
• Usable by the public (API or weights)
• Listed on an independent board we track
• Enough results for a rank in a category
• Holds across two snapshots (crowns only)

What the lab claims

Self-reported. Shown for context; not used in any verdict.

Lab claim: at 128K history, Nano delivers a 1.86× end-to-end inference speedup and about 66× lower persistent storage than a full-context KV cache.

Lab claim: across 11 memory benchmarks, MemT-1.0-Flash and MemT-1.0-Nano rank first and second among the evaluated parametric memory methods.

Sources