HOW TO RUN GLM-5.3 FLASH, DEEPSEEK V4 FLASH AND OTHER 321B FRONTIER MODEL ON YOUR OWN MACHINE 😳
with colibri repo. 34,227 stars. pure C, no GPU, no api key, no per token bill
what $0 gets you:
-GLM-5.3 Flash, 321B params, ~195 GB converted
-vision included
-DeepSeek V4 Flash, 284B, ~167 GB, or the REAP 150B cut at ~85 GB
-Kimi K3, 2.8T, ~1.6 TB snapshot, 32 GB+ RAM
-mac, linux and windows builds, no compiler needed
what this replaces:
-opus 4.8: $15/M in, $75/M out
-fable 5: $30/M out
-gpt-5.6 sol: $25/M out
-your monthly plan you forgot you're still on
all for $0
step 1: grab the release
download the release archive for Linux, macOS, or Windows.
unpack it, then run python3 coli info
nothing to install, no compiler, no pip
or git clone
github.com/JustVugg/colibri
cd colibri/c ./setup.sh cd ..
step 2: pick your model
GLM-5.3 Flash 321B, DeepSeek V4 Flash 284B, Kimi K3 2.8T
GLM needs a conversion pass, the other two load straight from huggingface
tight on disk? the REAP deepseek is 85 GB instead of 167
step 3: point it at the weights
python3 c/tools/convert_glm53.py --outdir /nvme/glm53_i4 --min-free-gb 30 # convert
GLM COLI_MODEL=/nvme/glm53_i4
./coli chat # start chatting
./coli web # live dashboard
./coli serve # headless api
step 4: tune it for your machine
./coli doctor # readiness check
./coli plan # ram/vram/disk placement
./coli tune # fastest safe profile
important: no GPU needed, a GPU only makes it faster. speed is set by your disk because the experts stream from it. 1.8 tok/s on a 128 GB desktop, 0.05 on a 25 GB dev box
the tradeoff is honest: no token bill, no hosted quota, but you need the storage
call me insane... but i cancelled my Elevenlabs subscription for this😳
a solo dev just open sourced the entire elevenlabs stack
29.8K stars on github and it runs 100% FREE on your own machine
voice cloning, video dubbing, dictation, audiobooks — 16 engines, 646 languages
no account. no api key. no usage meter.
you just need 8gb of ram and a gpu if you have one
here is how you set it up:
1. download 'VoiceStudio' from github releases (mac, windows, linux)
2. open it and load the default engine
3. drop a 10 second clip of your voice to clone it
4. type anything and hit generate
you will not find a FREE way to make AI Voices this year than this