Cactus Needle 3
Ranked 3 of 4 tool-calling models by recommenders
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.
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.
22 people gave an opinion
- recommend it
- 10
- criticize it
- 12
- approval share
- 45%
Recommended in 1 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: not recommended
- February 2026: not recommended
- March 2026: not recommended
- April 2026: not recommended
- May 2026: not recommended
- June 2026: not recommended
- July 2026: not recommended
- August 2026: not recommended
- September 2026: recommended
- October 2026: not recommended
- November 2026: not recommended (month in progress)
Why it doesn’t carry the badge
- 10 people recommend itNeeds 10 or more, met.
- 0.8 recommenders for every criticNeeds 2 to 1 or better, not met.
- Recommended in 1 of 13 monthsNeeds 4 or more, not met.
The same three rules apply to every product, and the badge can’t be bought. How the badge works
What people say about it
The reasons people give, with the number of comments that give each one. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Praised as innovative small-model work
“one of the most promising model due to its original architecture”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “I very much like to see innovation in this space, so well done!”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “I'm really happy that someone is building this.”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” Works on-device with limited compute
“I think this illustrates the intended deployment pretty well, where both computational resources and compute credits can both be issues for deployment.”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “The intelligence laddering and on-device tool calling are especially cool.”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” Adjustable scope and argument limits
“you are able to set explicit minimum and maximum values on the output range of numeric arguments”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “It could be useful if I change and limit the scope of the kinds of actions I need.”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” Good with direct commands
“needle shines through after putting in the work”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “but anything less direct didn't”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash”
What people warn about
Each mark is one comment
Poor results on real tasks
“Reasoning is often utter nonsense and the actions are very often very wrong.”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “it's dogshit at what people actually try to feed it”
Comment from 19 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “this model simply does not work at all”
Comment from 19 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” - Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash”
Too small for complex tasks
“could easily have access to a bigger model running on a more capable device”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “it's still not quite capable enough for my target compared to FunctionGemma”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “The model, while capable in many dimensions, is also limited by its size.”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” Misses or wrong tool calls
“At this point the vacuum is in a dark bathroom, the living room is 100% brightness and playing "wee".”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “the model actually turned the lights to a "warm brightness"”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “Only the last one actually used the emergency_sos function”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” Claims seem exaggerated
“The "DeepSeek 4 Flash grade" claim seems far fetched.”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “the copy on the landing page is clearly AI generated”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash” “for that task it was very bad MNLI was better”
Comment from 18 Sep 2026 in “Show HN: Cactus Needle 3: 8-29MB automation models can match DeepSeek V4 Flash”
Other tool-calling models people recommend
The three most recommended other tool-calling models, with the same counts for each, so you can compare like for like. The tick on each bar marks the 2 to 1 line the badge needs.
| Product | Recommend | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Needle2 | 51 | 22 | 70% | 5 | — | Approved |
| FunctionGemma | 11 | 2 | 85% | 3 | — | — |
| Cactus Needle 3This page | 10 | 12 | 45% | 1 | — | — |
| Needle 26M | 6 | 3 | 67% | 1 | — | — |
What people said
27 of the 30 comments that mention it, newest first. Each links to the original comment.
“this model simply does not work at all”
“it's dogshit at what people actually try to feed it”
“This is incredibly cool. I tried telling it to both turn up the temperature in the home and turn down the lights in the LR and got a beautiful JSON doc.”
“I very much like to see innovation in this space, so well done!”
“I think this illustrates the intended deployment pretty well, where both computational resources and compute credits can both be issues for deployment.”
“it's still not quite capable enough for my target compared to FunctionGemma”
“the model actually turned the lights to a "warm brightness"”
“Only the last one actually used the emergency_sos function”
“The intelligence laddering and on-device tool calling are especially cool.”
“the copy on the landing page is clearly AI generated”
“could easily have access to a bigger model running on a more capable device”
“one of the most promising model due to its original architecture”
“The "DeepSeek 4 Flash grade" claim seems far fetched.”
“It could be useful if I change and limit the scope of the kinds of actions I need.”
“Reasoning is often utter nonsense and the actions are very often very wrong.”
“The model, while capable in many dimensions, is also limited by its size.”
“for that task it was very bad MNLI was better”
“'crashed' implies need for music.”
“you are able to set explicit minimum and maximum values on the output range of numeric arguments”
“I wouldn't trust needle to be my autopilot either”
“needle shines through after putting in the work”
“A list of hardware platforms doesn't make a use case.”
“over that and it gets confused”
“I'm really happy that someone is building this.”
“but anything less direct didn't”
“At this point the vacuum is in a dark bathroom, the living room is 100% brightness and playing "wee".”
“None of the queries I asked worked”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.