Linux gaming is faster because Windows APIs are becoming Linux kernel features

Hey HN, Henry here from Cactus. We open-sourced Needle, a 26M parameter function-calling (tool use) model. It runs at 6000 tok&#x2F;s prefill and 1200 tok&#x2F;s decode on consumer devices.We were always frustrated by the little effort made towards building agentic models that run on budget phones, so we conducted investigations that led to an observation: agentic experiences are built upon tool calling, and massive models are overkill for it. Tool calling is fundamentally retrieval-and-assembly (match query to tool name, extract argument values, emit JSON), not reasoning. Cross-attention is the right primitive for this, and FFN parameters are wasted at this scale.Simple Attention Networks: the entire model is just attention and gating, no MLPs anywhere. Needle is an experimental run for single-shot function calling for consumer devices (phones, watches, glasses...).Training:
- Pretrained on 200B tokens across 16 TPU v6e (27 hours)
- Post-trained on 2B tokens of synthesized function-calling data (45 minutes)
- Dataset synthesized via Gemini with 15 tool categories (timers, messaging, navigation, smart home, etc.)You can test it right now and finetune on your Mac&#x2F;PC: <a href="https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;needle" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;needle</a>The full writeup on the architecture is here: <a href="https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;needle&#x2F;blob&#x2F;main&#x2F;docs&#x2F;simple_attention_networks.md" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;needle&#x2F;blob&#x2F;main&#x2F;docs&#x2F;simp...</a>We found that the &quot;no FFN&quot; finding generalizes beyond function calling to any task where the model has access to external structured knowledge (RAG, tool use, retrieval-augmented generation). The model doesn&#x27;t need to memorize facts in FFN weights if the facts are provided in the input. Experimental results to published.While it beats FunctionGemma-270M, Qwen-0.6B, Granite-350M, LFM2.5-350M on single-shot function calling, those models have more scope&#x2F;capacity and excel in conversational settings. We encourage you to test on your own tools via the playground and finetune accordingly.This is part of our broader work on Cactus (<a href="https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;cactus" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;cactus</a>), an inference engine built from scratch for mobile, wearables and custom hardware. We wrote about Cactus here previously: <a href="https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44524544">https:&#x2F;&#x2F;news.ycombinator.com&#x2F;item?id=44524544</a>Everything is MIT licensed. Weights: <a href="https:&#x2F;&#x2F;huggingface.co&#x2F;Cactus-Compute&#x2F;needle" rel="nofollow">https:&#x2F;&#x2F;huggingface.co&#x2F;Cactus-Compute&#x2F;needle</a>
GitHub: <a href="https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;needle" rel="nofollow">https:&#x2F;&#x2F;github.com&#x2F;cactus-compute&#x2F;needle</a>

Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model

Setting up a free *.city.state.us locality domain (2025)

A History of IDEs at Google

Claude for Small Business

The Emacsification of Software

Scorched Earth 2000 – Web

MacBook Neo Deep Dive: Benchmarks, Wafer Economics, and the 8GB Gamble

The US is winning the AI race where it matters most: commercialization

Chess puzzle I found in my dad's old book

A Claude Code and Codex Skill for Deliberate Skill Development

Classic 7 is a Windows 10 LTSC mod to look 1:1 to Windows 7

USDA Projects Smallest US Wheat Harvest Since 1972 Due to Plains Drought

Leaving the Physical World

Avoiding and reducing microplastic false positives from dry glove contact

An attempt at a single pass LLVM frontend in ~3000 lines of C without external dependencies, malloc, or an AST. Included are some graphical examples. The IR isn&#x27;t perfect, and the README touches on one particular downfall

Show HN: Nibble

Every privacy-focused DNS service requires an account: NextDNS, Cloudflare for Families, Apple&#x27;s iCloud Private Relay (paid, iOS-only). The protocol that doesn’t require one - ODoH - had basically one well-known public relay operator (Frank Denis on Fastly Compute, default in dnscrypt-proxy). I built a second one and the client to talk to it.

Show HN: Running the second public ODoH relay

Computer Hobby Movement in Canada

The Siri for Families Apple Will Never Build

Sam Altman's Business Dealings Under GOP Scrutiny Ahead of OpenAI's IPO