๐Ÿ–ฅ๏ธ Free AI on any CPU โ€” three values, no graphics card (BitNet)

Download one 1.19 GB file, open it on an ordinary laptop, and it answers you โ€” no graphics card, no account, nothing to rent.

Microsoftโ€™s BitNet stores every brain-weight as one of three values: โˆ’1, 0, +1. A tiny lookup turns those values into words using only addition. Thatโ€™s why the file is smaller than most phone games.

digraph G { graph [bgcolor="transparent"]; node [shape=box, style="rounded,filled", fontname="Helvetica", fontsize=15, margin="0.22,0.13", penwidth=0]; edge [penwidth=1.5, color="#8b93a1", arrowsize=0.7, fontsize=12]; a [label="๐Ÿ‘€ Just looking", fillcolor="#FFF1DC"]; b [label="๐Ÿ’ป Laptop in front of me", fillcolor="#E7F0FB"]; c [label="๐Ÿง  I want the real model", fillcolor="#E6F5EC"]; d [label="โ˜๏ธ No machine today", fillcolor="#EFEAFB"]; z [label="โœ… AI answers on your terms", fillcolor="#DDF2E4"]; a -> z [label="open a demo"]; b -> z [label="one-line install"]; c -> z [label="official 1.19 GB model"]; d -> z [label="free daily compute"]; }

What you actually get

:desktop_computer: BitNet โ€” Microsoftโ€™s 1-bit AI engine, MIT-licensed, ~42k stars.

:package: The model you can download today: 2.4B parameters, 1.19 GB โ€” about 1.5 GB of RAM while it runs.

:high_voltage: Measured: 1.37โ€“5.07ร— faster on Apple CPUs, 2.37โ€“6.17ร— on Intel, with 55โ€“82% less energy than ordinary 16-bit AI.

:key: The win is the map, not the model: three values plus a lookup beat stacking more graphics cards.

See it before you install

:globe_with_meridians: Official live demo โ€” type a question, get an answer.

:teddy_bear: 2.31 MB toy โ€” 11 million parameters, and you can watch its brain as โˆ’1 / 0 / +1 signs.

:video_game: 0xBitNet chat โ€” the full 2B model runs inside your browser; nothing leaves your machine.

:bar_chart: Speed scoreboard โ€” real tokens-per-second numbers on a cheap machine.

Put it on your laptop

:high_voltage: BitRun โ€” one install line, prebuilt files, a speed dashboard, and a 79 MB starter model.

:hammer_and_wrench: Official build โ€” clone the repo, then:

git clone --recursive https://github.com/microsoft/BitNet.git
cd BitNet
conda create -n bitnet-cpp python=3.10 && conda activate bitnet-cpp
pip install -r requirements.txt
huggingface-cli download microsoft/BitNet-b1.58-2B-4T-gguf --local-dir models/BitNet-b1.58-2B-4T
python setup_env.py -md models/BitNet-b1.58-2B-4T -q i2_s

:electric_plug: Fast mode needs AVX2 on x86, or NEON on ARM.

Push it further

:1st_place_medal: FairyFuse โ€” same three values, a kernel that only adds and subtracts: 29.6ร— faster kernel, 32.4 tokens/s on one Intel server chip. Authorsโ€™ measurements, early paper.

:2nd_place_medal: Litespark โ€” one pip install, then load a real 1-bit model; up to 18ร— throughput on Apple silicon. Early paper.

:3rd_place_medal: T-MAC โ€” the lookup trick now inside bitnet.cpp: 3.4ร— on an M2 Ultra, 5.8ร— on a Raspberry Pi 5. Peer-reviewed.

:whale: DeepSeek-R1 671B โ€” a real 671-billion-parameter model packed into 131 GB; works on ~20 GB of RAM, slowly.

:test_tube: LittleBit โ€” research pack putting Llama2-13B under 0.9 GB. Early research, quality drops.

Bigger than your RAM

:floppy_disk: SSD offload โ€” a 25 GB model on a 16 GB laptop with two extra launch flags.

:electric_plug: PocketLLM โ€” stream a 30B model off a USB stick in 6 GB of RAM. New project.

:abacus: solid.cpp โ€” a one-line patch that lets a 93 GB model page from disk on a 42 GB machine. New project.

:mobile_phone: DistAi โ€” pool three old phones over USB into one 14B machine; keep it on a private network.

:high_voltage: ik_llama.cpp โ€” one -rtr 1 flag gave up to 7ร— faster prompt reading on the same CPU.

No machine? Use free compute

:cloud: Cloudflare Workers AI โ€” 10,000 free units a day, roughly 49,000 output tokens on a 70B model, no card.

:test_tube: GitHub Models โ€” your GitHub login is the key; 40+ models from an Action.

:puzzle_piece: Hugging Face thin Space โ€” host a big-model demo that bills the visitor instead of your free quota.

:desktop_computer: Kaggle โ€” 30 h/week of free 32 GB graphics cards ยท GitHub Actions โ€” free CPUs you can fan out.

Read once before you start

:straight_ruler: The โ€œ100-billionโ€ line going around the web is a benchmark setting in the paper, not a download. The model you can actually get is 2.4B / 1.19 GB; that benchmark ran 6.58 tokens/s on a Mac Studio, 1.70 on an Intel laptop (source).

:wrench: The official source has a one-line activation fix in PR #642 โ€” applying it restores the intended answer quality (perplexity 17.1 instead of 99.8). Windows builders are tracking PR #644.

:mobile_phone: For loading the model yourself, use bitnet.cpp or the browser demos โ€” app support is still arriving (request).

:test_tube: FairyFuse and Litespark are early papers โ€” the numbers are the authorsโ€™ own measurements.

:cloud: Free compute tiers are for your own experiments; keep them personal.