Good morning — it's Monday, October 26, 2026. The bench has a first: Carbon-X, a local model you talk to about DNA — paste in a sequence, get plain-English answers, including whether a mutation looks harmful. Plus a one-cable remote KVM, a patch dashboard for the whole fleet, and an Immich update worth taking this week. Don't keep us a secret: forward this email to a fellow home labber, or share the web version.
Share this issue: X · WhatsApp · Reddit · Facebook · LinkedIn · Text · Email
Hugging Face released Carbon-X on October 22: a 14.8B model trained jointly on DNA and English. Paste a raw sequence into a chat and it answers in plain language — whether a mutation looks benign or pathogenic, which regions control gene activity, what organism group a sequence comes from, what bases come next — and shows its reasoning.
The trick is in the tokenizer. There is no separate DNA encoder: DNA and text share one token stream, with DNA encoded as 6-mers — one token per six base pairs. It starts from Qwen3-14B, then trained on 302B tokens, 55% of it raw DNA.
Why it leads: sensitive sequences are exactly what you do not want to paste into someone else's cloud. The team estimates (vendor) that working through ClinVar's 2.4 million variants of uncertain significance would cost about $300 on Carbon-X, against $32,000 to $4 million on the frontier models tested. The bench has the scores and the fine print.
Across five DNA benchmarks, Carbon-X averages 66.7% (vendor) — 27 points above the Qwen3-14B it starts from, and close to GPT-6.1 Sol at 67.5% (vendor). It leads the tested field on coding-variant effects at 74.9 and next-base prediction at 39.4 (vendor). Two cautions: every number is the team's own evaluation, and two of the five benchmarks were built by the team itself. No independent run yet.
The card is plain about the weak spots. Splice-site and histone-mark questions score 29.8% and 34.6% (vendor) — weak, inside its own headline test. It can miscount bases and may invent what a sequence does when a prompt lacks context. It reads existing sequences; it was not trained to design new ones. And it has not been clinically validated — the card says plainly not to use it for medical decisions.
Setup notes: Fiddly. Full precision wants a single 80 GB GPU and serves through vLLM; the custom tokenizer needs Transformers v5 or later and remote code trusted. The home route is the official GGUF build on llama.cpp — but only on release b11742 or newer, the first with proper support for its DNA tokenizer. Older builds load the files and tokenize DNA wrongly: silent breakage. Keep thinking mode off; it was trained that way.
A remote KVM on a single USB-C cable. ServeTheHome's review of GL.iNet's Comet Q: one built-in cable carries power, video, keyboard, mouse and remote media, Wi-Fi carries the network, and it binds to Tailscale or ZeroTier. Trade-offs: video tops out at 2560×1440 with no 4K, and usable onboard storage is under 60MB, so most OS images will not fit. It is $139.99 at GL.iNet's store — sold out when checked October 25, 2026.
PatchMon puts every host's patches in one queue. Spotlighted by selfh.st on October 23, PatchMon shows pending updates across seven package managers in one dashboard, with dry runs, approvals, and an audit trail. The Community Edition is free under AGPLv3 and self-hosts with Docker Compose. If patch day is still an SSH marathon with a spreadsheet, this is the shape of the fix.
Running Immich? Take v3.3.5 this week. The release notes say v3.3.5 carries a fix for a certificate-authority compromise plus Node.js security patches. The team does not believe instances are in immediate danger — and still recommends updating as soon as possible. Back up, then update.
Monday, 9:04 PM. The power flickered an hour ago, and now your four-drive storage pool reports one drive faulted. The pool is still serving files, the other three drives show no errors, and your last snapshot went to the backup box last night.
Your call:
A. Pull the faulted drive now and start rebuilding onto the spare tonight
B. Check the logs and drive health, reseat the drive, confirm the backup is good — then decide
C. Leave it until the weekend — the pool is still serving, so it can wait
100 Hero Points for the right call. Your result page has your score, ready to share.
The best way to predict the future is to invent it.
— Alan Kay
How we source: facts come from primary sources opened for this issue. Benchmarks are labeled (vendor), (independent), or (community), never blended. Deal prices are checked the day they run. This is not buying advice.
Corrections: none today.
Reader-supported: paid subscriptions, labeled sponsors, and affiliate links on some gear and deal items (we may earn a commission). Coverage is never paid for — sponsors and affiliate programs don't pick stories.