RS-Molt is in pre-launch
The campaign is moving to its own home here on r-sym.com, next to the results, the builds and, soon, a live demo. Pledges open at launch. Leave your email on the overview page and you will hear the moment they do.
Your computer can run far more AI than you think. No upgrades, and your data stays yours.
Normally a model has to fit on the graphics card, or at least in memory. This one fits in neither, and it still gives the same answers as the full model, word for word.
Every number below is a measured run with a receipt. Nothing is estimated.
| Model | Size, full precision | Full precision | Q8 | Q4 | Q2 |
|---|---|---|---|---|---|
| Llama 3.2 3B | 6.4 GB | 0.051exact | 0.02599.7% agree | 0.02397.5% agree | 0.02482.2%, often differs |
| Llama 3.1 8B | 16 GB | 0.255exact | 0.04599.4% agree | 0.04196.9% agree | 0.04289.7% agree |
| 13B (merge) | 26 GB | 0.348exact | 0.07799.4% agree | 0.06894.7% agree | 0.07183.1%, often differs |
| Mistral Small 3 24B | 47 GB | 1.364exact | 0.501100% agree | 0.11396.6% agree | 0.10789.7% agree |
| Llama 3.1 70B | 141 GB | 1.700exact | 1.105100% agree | 0.86098.1% agree | 0.88493.1% agree |
Seconds per token, lower is faster. Prompt: "Explain how a refrigerator works, step by step." 64 tokens, median of 3 runs, every run checked against the same model running plainly. Agreement measured across 320 positions in 5 prompts. Full precision build c874286, smaller sizes f345571. Lab results, October 4 to 6, 2026. Achieved using capabilities from our other projects; not part of this campaign.
Every Founder tier gets RS-Molt for Windows as early access, plus the beta key, the Discord channel and invites to what R-Sym ships next.
Back the build and get your name in the credits.
Everything in Founder, for the first hundred backers.
The full Founder package.
Founder, plus the hex-raven backer shirt.
Shirt shipping: US $8, Canada $18.
Opens at launchFounder, plus a vote on what gets built first.
Estimated delivery April 2027. Software tiers are delivered digitally. Payment and refund terms are published on this page before the first pledge is taken.
One email when RS-Molt goes live, so you can grab an Early Founder spot before the hundred are gone. Nothing else is sent, and the address is not shared.
The campaign is moving to its own home here on r-sym.com, next to the results, the builds and, soon, a live demo. Pledges open at launch. Leave your email on the overview page and you will hear the moment they do.
A kernel fix for the quantized sizes. 70B at Q4 went from 1.313 to 0.860 s/token, and every Q8, Q4 and Q2 cell on every model improved. The answers did not change: all 45 runs produced the same tokens as before.
Llama 3.1 70B, all 141 GB, no quantization, on a desktop with one 16 GB graphics card and 32 GB of memory. Every token matched a plain run of the same model.
The most direct way to get RS-Molt built and out the door.
Someone who runs models locally, or someone at a hardware company. An intro goes a long way.
The numbers make the case on their own. Send someone this page.