by Google
Google research training framework alternating wake and sleep phases for memory consolidation and self-generated practice.
Updated EU status checked Last shown
Sources: creator videos at the linked timestamps. EU status is information, not legal advice; see how we check.
Sleep (also referred to as the Sleep framework) is a research training framework attributed to Google. According to a single creator summary, it splits training into wake and sleep phases. During the sleep phase the model consolidates memory and generates its own practice problems.
The same source reports near-perfect long-context memory at up to 10M tokens, and describes the work as an early prototype. No official page, paper or code could be retrieved, so these claims are unverified and details such as architecture, benchmarks and release date are unknown.
There is no known public access route. It appears to be research-only and not a product users can try.
Watch the moment: the exact second top creators show it. Videos load only when you press play, via youtube-nocookie.com.
Training framework with wake and sleep phases where the model consolidates memory and generates its own practice problems. Near-perfect long-context memory up to 10M tokens; early prototype.
Closest neighbours by meaning, not just by category.
Model · MemTensor
Open-weights memory foundation model from MemTensor with a fixed-size persistent memory state, built on Qwen3.6-35B-A3B.
Research
Research framework letting a frozen VLM agent learn robot tasks in context from videos, images and feedback, no fine-tuning.
Model · Google
Google's open-weights 26B diffusion language model that drafts blocks of text in parallel.
Research · Google
Google research technical report on improving an agent's exploration policy by replaying past discovery trees offline.
Research · Stanford Hazy Research
Research language-modeling approach that generates text coarse-to-fine, producing tokens in parallel within each stage.
Research · DeepSeek
DeepSeek research on a separate large conditional memory module for static facts, held in server RAM instead of GPU memory.