LittleBit: sub-1-bit LLM compression via latent factorization
Samsung Research method compressing LLMs to 0.1 bits per weight via latent matrix factorization plus binarization (31x memory cut), with official code including the LittleBit-2 follow-up.
Why it matters
Pushes quantization to 0.1 bits per weight while keeping the model architecture unchanged at inference, a regime where previous methods broke down.
What you could build with it
A mobile app developer could compress a 13B-class model to under 1GB with LittleBit and ship a fully offline document assistant inside an app, removing server inference costs and working without connectivity.
Does it hold up?
Peer-reviewed method (NeurIPS 2025, ICML 2026 follow-up) with strong reported numbers, but independent reproduction reports from practitioners are not yet visible — too early to judge real-world reliability.
Built with LittleBit
- Samsung's LittleBit squeezes AI models to a tenth of a bit per weightstartupfortune · Third-party writeup 17h ago: walks through the latent-factorization method, the 31x memory cut (Llama2-13B under 0.9GB), and the NeurIPS 2025 / ICML 2026 paper trail.
Learn more
First spotted on github: source.
More AI models
OpenAI publishes 722 AI-generated math manuscripts from an unreleased modelOpenAI released 722 mathematical manuscripts in 372 result families to GitHub under Apache-2.0, produced by an…models · JEV 0.74Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.