# Cactus Needle 3 10 Hacker News commenters recommended Cactus Needle 3 and 12 criticized it, November 2025 to November 2026. It was recommended in 1 of 13 months. **Verdict:** A small on-device model that turns requests into structured tool calls, praised for its original architecture and explicit argument limits; critics say it is too limited for complex tasks and often gets actions wrong. ## Why people recommend it - Praised as innovative small-model work (3 comments): "one of the most promising model due to its original architecture" ([source](https://news.ycombinator.com/item?id=49759379)) "I very much like to see innovation in this space, so well done!" ([source](https://news.ycombinator.com/item?id=49760889)) "I'm really happy that someone is building this." ([source](https://news.ycombinator.com/item?id=49757442)) - Works on-device with limited compute (2 comments): "I think this illustrates the intended deployment pretty well, where both computational resources and compute credits can both be issues for deployment." ([source](https://news.ycombinator.com/item?id=49760535)) "The intelligence laddering and on-device tool calling are especially cool." ([source](https://news.ycombinator.com/item?id=49759737)) - Adjustable scope and argument limits (2 comments): "you are able to set explicit minimum and maximum values on the output range of numeric arguments" ([source](https://news.ycombinator.com/item?id=49758178)) "It could be useful if I change and limit the scope of the kinds of actions I need." ([source](https://news.ycombinator.com/item?id=49759084)) - Good with direct commands (2 comments): "needle shines through after putting in the work" ([source](https://news.ycombinator.com/item?id=49758054)) "but anything less direct didn't" ([source](https://news.ycombinator.com/item?id=49757288)) ## What people warn about - Poor results on real tasks (4 comments): "Reasoning is often utter nonsense and the actions are very often very wrong." ([source](https://news.ycombinator.com/item?id=49759060)) "it's dogshit at what people actually try to feed it" ([source](https://news.ycombinator.com/item?id=49763383)) "this model simply does not work at all" ([source](https://news.ycombinator.com/item?id=49769194)) - Too small for complex tasks (3 comments): "could easily have access to a bigger model running on a more capable device" ([source](https://news.ycombinator.com/item?id=49759380)) "it's still not quite capable enough for my target compared to FunctionGemma" ([source](https://news.ycombinator.com/item?id=49760514)) "The model, while capable in many dimensions, is also limited by its size." ([source](https://news.ycombinator.com/item?id=49758900)) - Misses or wrong tool calls (3 comments): "At this point the vacuum is in a dark bathroom, the living room is 100% brightness and playing "wee"." ([source](https://news.ycombinator.com/item?id=49757107)) "the model actually turned the lights to a "warm brightness"" ([source](https://news.ycombinator.com/item?id=49760339)) "Only the last one actually used the emergency_sos function" ([source](https://news.ycombinator.com/item?id=49759897)) - Claims seem exaggerated (3 comments): "The "DeepSeek 4 Flash grade" claim seems far fetched." ([source](https://news.ycombinator.com/item?id=49759153)) "the copy on the landing page is clearly AI generated" ([source](https://news.ycombinator.com/item?id=49759614)) "for that task it was very bad MNLI was better" ([source](https://news.ycombinator.com/item?id=49758394)) ## Alternatives hackers recommend - [Needle2](https://hackerapproved.com/p/needle2.md): 51 recommend - [FunctionGemma](https://hackerapproved.com/p/functiongemma-270m.md): 11 recommend - [Needle 26M](https://hackerapproved.com/p/needle-26m-tool-calling-model.md): 6 recommend ## Cite as Hacker Approved, "Cactus Needle 3", https://hackerapproved.com/p/cactus-needle-3/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).