hugging face released Carbon-X: paste raw DNA into a chat and a 14b model answers in english, benign or pathogenic, with reasoning. within a point of the frontier model at a tiny fraction of the cost, and it fits on one gpu so your sequence stays yours. okay this is cool
Today we're releasing Carbon-X, a 14B model that brings DNA and natural language together for the first time 🧬
1. Beats every open language model we tested across DNA benchmarks. Within a point of GPT-6.1 Sol at >1,000× lower inference cost.
2. Mid-trained on 300B tokens of interleaved DNA and text, followed by SFT (without distillation!) and RL.
3. A new hybrid tokenizer encodes DNA using ~3× fewer tokens than BPE.
4. Gives other models a DNA specialist subagent: Kimi K3 gains 11.1 points when it consults Carbon-X.
5. Runs locally on a laptop with llama.cpp. Apache 2.0. Your DNA stays on your hardware.
Model: huggingface.co/HuggingFaceBi…
Blog: huggingface.co/spaces/Huggin…
ant group's ling team shipped another flash model, 44 on the index people rank these things with, up from 41, and the agent jumps are the actual story. the open weights aren't out yet and that's the part i'm waiting on
bun just baked a prettier into the runtime. it formats prettier's own repo in 60ms, prettier takes 17s, and you delete a dev dependency instead of adding one. nice work team
Bun v1.4.4
- Fixes 271 issues, addressing 544 👍
- `bun fmt` is a fast Prettier-compatible code formatter
- `bun check` gets 1.4x - 7.8x faster
- New in Bun.SQL & WebSocket
- Security fixes, including for Brotli, TLS certs, HTTP/3, and more
bun.com/blog/bun-v1.4.4
an anonymous model called Kibitz showed up on LMArena's code arena and testers are reading it as Kimi's next one sneaking out early. one prompt got an interactive pelican-on-a-bike page with working pedals and a night mode. if the real release lands like this, sheesh
higgsfield shipped content factory: you type one "what if" idea, pick a trending style, out comes a whole series of ready-to-post vertical clips. the whole feed is about to be the same six videos and i will watch all of them
nvidia dropped open embed models at GTC berlin and turbopuffer already has them: your docs get turned into search vectors as you write, no embedding provider to wire up. that whole plumbing job gone
Nemotron-3.5-Embed is now available in turbopuffer native embedding
on our enterprise PDF search benchmark, Nemotron-3.5-Embed 8B improves recall@20 by ~5% over the next best model
turbopuffer.com/docs/embeddi…
sebastien at foxio found that one HTTP/3 connection, no login, can eat a server's memory until it drops. fix is out for quic-go and quiche (so caddy, traefik, cloudflared), and turn HTTP/3 off on bun, mitmproxy and the rust h3 crate until theirs lands
‼️ A single HTTP/3 connection can crash vulnerable web servers. No login required.
Newly disclosed QPACK flaws trigger up to 4,000× memory amplification in tests, exhausting server resources.
→ PoC is public
→ 7 implementations affected
→ Still unpatched: Bun, aioquic, h3
Full details ↓ thehackernews.com/2026/10/ne…
new in vscode: one prompt, several different AI models running it at once, answers side by side on your own code, plus a judge scoring them. a leaderboard can't tell you that
🚀 The latest @code release makes the Copilot SDK-based harness the default for new sessions, adds model comparisons, and brings a file list to multi-file diffs.
🧵 Here’s what’s new:
code.visualstudio.com/update…
vercel's token pass now includes deepseek's new flash model: $30/mo buys your coding agent $80 of tokens to burn. top reply is already saying somebody else does the same for $10
DeepSeek V4.1 Flash 1017 is now included in Vercel Token Pass.
• New weights from DeepSeek
• 1M context window
• No config change
Get $80 of tokens for $30/mo ↓vercel.com/changelog/deepsee…
amp now lets you point its coding agent's disposable machines at your own boxes. docker, k8s, apple containers, whatever you've got. free for individuals and teams. a launch page saying "you probably shouldn't self-host this" is the most honest thing i've read
walmart is letting AI agents actually check out for you, each with its own profile and a spending cap you set. amazon won't even let one through the door
As Amazon blocks AI agents from shopping on its site, Walmart is taking the opposite approach, introducing new technology that lets AI agents make purchases on customers' behalf, with controls over spending and transaction approvals.
spr.ly/6019BHXyoW
deepseek just raised ~$15b, one of china's biggest rounds ever, with tencent and catl in. the lab that made cheap training look easy finally has a war chest to buy real compute
The Chinese AI pioneer DeepSeek has raised about $15 billion in its latest fundraising round, one of the country’s largest ever and a sign of the startup’s ambitions to compete globally with the likes of Anthropic and OpenAI bloomberg.com/news/articles/…
google put gemini inside google maps on carplay, first voice assistant maps has ever had there. mid-drive you just say "avoid tolls" or "find parking" and it reroutes. okay this one i'd actually use
Gemini is now available in @GoogleMaps navigation on CarPlay. Find the best place to park, report a traffic incident, or change your route with just a quick conversation.
Tap the Gemini icon in Google Maps CarPlay during your next drive to try it out 🚘
nvidia's robot world models are live on reactor now: you drag a robot's arms around in the browser and it generates what that robot would see next, in real time. rehearsing a robot before you burn hardware, finally
Introducing NVIDIA Cosmos-3-Nano on Reactor. Two new models, available now via the Reactor API.
• Sim-Bimanual predicts what a robot sees as you move its arms and grippers.
• Sim-Transfer turns simulations into photorealistic video.
Try them now: reactor.inc/cosmos-bimanual
nvidia's desk box now comes in 64GB, five grand from acer/asus/dell/hp/msi, shipping friday. chain two together and a real model runs at home on 128GB, no cloud, and the launcher sets your coding agent up for you
hedera wired their chain into one MCP endpoint: point Claude Code or Codex at it and the agent sends HBAR and mints tokens on testnet, but nothing settles until you sign from your own wallet. agents with wallets are coming, this one at least can't spend without me
The Hedera Network MCP is live on testnet. Connect Claude Code, Codex, Cursor, ChatGPT & more with one endpoint, then ask your agent to send HBAR, mint tokens, or query the network. Nothing settles until you sign it in your own wallet.
Try the Hedera Network MCP: hedera.com/mcp-servers/
