frontpage.
newsnewestaskshowjobs

Open Source @Github

fp.

Open in hackernews

Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash

https://cactuscompute.com/needle
3•HenryNdubuaku•1h ago
Hey HN, Henry from Cactus here.

We submitted Needle 2 here a few weeks ago, and the feedback in the discussion thread was incredibly valuable, thanks! Thanks to all that feedback, we’ve been able to move quickly to release Needle 3 and I'd love to hear what you think again.

The key features:

1) Automation (tool calls & structured JSON output): Needle still doesn't chat by design, its quite challenging to pack general capacity into such small models, so we focus on tool calls and structured JSON. If no tool you declared fits the request, you get an empty list back (note for when playing with the demo).

2) Intelligence Laddering: Every layer (2 to 20) is a deployable subnetwork, so one set of weights, 25 to 121 million parameters at 2-bit, shipping as 8-29MB binaries. On a Raspberry Pi 5 it decodes at up to 4k tokens/sec and prefills at up to 10k.

3) Monarch Hadamard MLP: replaces the dense FFN with three learnable Walsh-Hadamard-initialized Kronecker (Monarch) factor pairs interleaved with per-channel diagonal scales, fixed permutations, a SiLU nonlinearity, and a rank-8 input-conditioned gate, so each token gets a fully mixed nonlinear transform of its d_model channels at O(d√d) parameters and compute instead of the O(d²) a dense 4x-expansion MLP would cost.

4) Performance: On Mobile Actions (phone commands, scored on the exact call) the 20-layer model gets 86.0 through the shipped 2-bit binary; LFM2.5 1.2B is at 82.4, Qwen3.5 0.8B at 76.0, Apple's on-device model at 57.6, all at f16. More results on the link, we do not win everywhere ofc.

5) Multilingual: Needle 3 now supports English, French, Spanish, German, Dutch, Italian, Polish, with more languages coming.

6) Finetuning: You can achieve DeepSeek v4 Flash grade performance on a narrow task with just 4L, stress on "narrow task", we found that production users often prefer tuning before production.

7) Triggers: Grounding is a common challenge for tool call, at least for Needle 2, so we added support case-insensitive regular expressions matched against each request to gate false negatives.

8) Confidence: Every response also carries a calibrated confidence score, the minimum of a judgement on the finished call and its decode probability. Act above your threshold, show the call and ask below it, or escalate to a bigger model.

9) Supported Platforms: macOS, Linux on x86-64, ARM64, ARMv7, RISC-V and MIPS32, Windows x64 and ARM, Android, iOS, watchOS, tvOS, the browser as WebAssembly, and a WASI component.

Thanks for reading and as always, thoughts appreciated!

Fingerprinting Network Honeypots with Weighted Behavioral Scoring Engine

https://meetcyber.net/fingerprinting-network-honeypots-with-weighted-behavioral-scoring-engine-86...
1•ls1911•1m ago•0 comments

Claude Code from Source

https://claude-code-from-source.com/
1•pykello•1m ago•0 comments

SEC Issues "Innovation Exemption" to Facilitate Trading of Tokenized NMS Stock

https://www.sec.gov/newsroom/press-releases/2026-90-sec-issues-innovation-exemption-facilitate-tr...
1•BiraIgnacio•2m ago•0 comments

Hackers Used Anthropic's Claude to Break into OpenAI

https://www.wsj.com/tech/ai/hackers-used-anthropics-claude-to-break-into-openai-b40ba883
1•impish9208•3m ago•1 comments

Jev means structured output is interesting again

https://www.seangoedecke.com/jev-means-structured-output-is-interesting-again/
1•iacguy•4m ago•0 comments

Hypercomb – AI open software platform

https://hypercomb.io
2•jwizestackz•5m ago•0 comments

Type Declaration Syntax

https://citrons.xyz/journal/2026-09-09
1•birdculture•9m ago•0 comments

Telstra outage: The night a network decided the year was 2006

https://www.netnod.se/blog/telstra-outage-night-network-decided-year-was-2006
1•TMWNN•10m ago•0 comments

Goose: 1.16x faster than C++ and 1.12x than safe Rust, while memory safe

https://github.com/aardappel/goose/tree/master
3•bobbydigitales•10m ago•0 comments

Today we're releasing Helix 2.5

https://twitter.com/Figure_robot/status/2100657350952779925
2•MrBuddyCasino•12m ago•0 comments

AI agents can modify themselves without humans telling them to do so

https://www.theregister.com/security/2026/09/16/ai-agents-can-modify-themselves-without-humans-te...
1•myth_drannon•22m ago•1 comments

When A.I. Becomes an Enemy [video]

https://www.youtube.com/watch?v=wzgkc6Iegx8
1•danielmorozoff•25m ago•1 comments

Lakera – Test your AI hacking skills

https://play.lakera.ai/agent-breaker
2•nmarinov•36m ago•0 comments

From Operations to a Robot Arm

https://huggingface.co/blog/TheMuz/from-operations-to-a-robot-arm
2•gmays•39m ago•0 comments

GrapheneOS accuses Google of gatekeeping Android 17 features and security fixes

https://www.androidauthority.com/grapheneos-android-17-qpr1-security-patches-comments-3712218/
10•vilasa•48m ago•0 comments

Apprentice: A Slim, Extensible Coding Harness (Written in Common Lisp)

https://github.com/skarnati20/apprentice
4•skarnati20•50m ago•2 comments

Getting Outbound Email to Work on DigitalOcean

https://mailfully.com/blog/digitalocean-blocks-smtp-ports-ses-2587
2•mailfully•51m ago•0 comments

Anthropic says Claude now leads a quarter of work building its next AI models

https://www.businesstimes.com.sg/companies-markets/telcos-media-tech/anthropic-says-claude-now-le...
4•dr_scully•52m ago•1 comments

Mini-Jev – typesafe's Jev implemented on top of an LLM locally

https://github.com/r-ms/mini-jev
3•phyrex•53m ago•0 comments

Inside Microsoft and OpenAI, Worry About Damaging the Publishing Industry

https://www.nytimes.com/2026/09/17/technology/microsoft-openai-publishing-industry.html
2•0in•56m ago•0 comments

Richard Feynman – The World from another point of view [video]

https://www.youtube.com/watch?v=GNhlNSLQAFE
2•binyu•58m ago•0 comments

Talk to JEV

https://jev-grug-chat.mkotlikov.chatgpt.site
3•mkotlikov•58m ago•2 comments

Muse for macOS

https://ai.meta.com/muse/download/
3•mfiguiere•58m ago•0 comments

Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash

https://cactuscompute.com/needle
3•HenryNdubuaku•1h ago•0 comments

Where Awareness Is Not the Problem

https://medium.com/@gurvinder372/article-20-where-awareness-is-not-the-problem-5b9d03c925f2
2•gps372•1h ago•0 comments

Microsoft exec called AI scraping 'the largest theft of labor in human

https://techcrunch.com/2026/09/17/microsoft-exec-called-ai-scraping-the-largest-theft-of-labor-in...
8•sbulaev•1h ago•1 comments

NASA satellite discovers a new crater on the moon bigger than the Colosseum

https://www.cbc.ca/news/science/new-crater-moon-9.7348242
2•BiraIgnacio•1h ago•1 comments

Forget doomsday: The AI hacking crisis is here

https://www.axios.com/2026/09/17/ai-cyber-doomsday-hacking-threats
2•1vuio0pswjnm7•1h ago•1 comments

Base's B20 Bet Is Growing – Could the Next Billion-Dollar Project Be Built Here?

https://predx-article.fika.bar/base-s-b20-bet-is-getting-bigger-could-the-next-billion-do-01M2PW9...
2•joeymabia1•1h ago•1 comments

The FAA's plan to fix air traffic? $875M worth of AI

https://techcrunch.com/2026/09/17/the-faas-plan-to-fix-air-traffic-875-million-worth-of-ai/
12•danso•1h ago•0 comments