Needle2
Hacker ApprovedThe most recommended of 4 tool-calling models
Tiny 26M-parameter model distilled from Gemini that converts simple instructions into tool calls on phones, wearables and embedded devices. Critics say it handles only simple, templated requests and reasons poorly.
Tiny 26M-parameter model distilled from Gemini that converts simple instructions into tool calls on phones, wearables and embedded devices. Critics say it handles only simple, templated requests and reasons poorly.
73 people gave an opinion
- recommend it
- 51
- criticize it
- 22
- approval share
- 70%
Recommended in 5 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: recommended
- February 2026: not recommended
- March 2026: not recommended
- April 2026: not recommended
- May 2026: recommended
- June 2026: not recommended
- July 2026: recommended
- August 2026: recommended
- September 2026: not recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it carries the badge
- 51 people recommend itNeeds 10 or more, met.
- 2.3 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 5 of 13 monthsNeeds 4 or more, 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
Handles simple tool calls well
“Using agent for simple tasks is inefficient and wasteful, Needle really resolves this.”
Comment from 13 May 2026 in “Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model” “It seems to convert simple instructions to reasonable tool calls.”
Comment from 10 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “it did a pretty good job for tool calling when I tried it”
Comment from 10 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” - Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots”
Small, cheap alternative to large agents
“I like Needle a lot, and I love the idea of any tiny resource-thrifty LLM that can run on older hardware, that outputs only JSON!”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “they're incredibly cheap and handle NLP (not in a predictable way of course, but with great success rates)”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “this could be a nice model to have as the first pass in a more complex agent system”
Comment from 13 May 2026 in “Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model” Runs on tiny embedded devices
“This is a way that everyone can get that on the actual Apple Watch or local phone”
Comment from 12 May 2026 in “Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model” “being able to do that on a small embedded device is the novelty here”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “i made it run on esp32s3, and it works!!!”
Comment from 20 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” Fits Home Assistant use cases
“This is pretty much exactly what I want for Home Assistant.”
Comment from 12 May 2026 in “Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model” “This would be amazing for home assistant.”
Comment from 12 May 2026 in “Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model”
What people warn about
Each mark is one comment
Weak reasoning and general knowledge
“It seems to me that the model struggles to have enough general intelligence, knowledge, or reasoning capacity”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “I don't think consumers would consider this "intelligence". It's too dumb.”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “models of this size can't reason about relative values very well”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” - Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots”
Fails outside a few templates
“the web demo is not particularly impressive. It really doesn't like anything I throw at it.”
Comment from 10 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “Since it seems limited to matching a few templates and otherwise falling flat on its face”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “limited to matching a few templates and otherwise falling flat on its face”
Comment from 10 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” Wrong or ignored tool arguments
“to set the mode literally the opposite of what was requested.”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “it just completely ignored the "brightness" parameter”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” Output quality often too low
“if the number of cases where it works as expected is lower than what we could catch with a (old-Siri-style) heuristics based approach”
Comment from 11 Aug 2026 in “Show HN: Needle2: 14MB agentic LLM for phones, wearables, smart home and robots” “Context definitely helps. But yeah the quality of it doesn't seem to be too high.”
Comment from 13 May 2026 in “Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model” “Which is kind of a little way there but not useable on its own”
Comment from 14 May 2026 in “Show HN: Needle: We Distilled Gemini Tool Calling into a 26M Model”
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 |
|---|---|---|---|---|---|---|
| Needle2This page | 51 | 22 | 70% | 5 | — | Approved |
| FunctionGemma | 11 | 2 | 85% | 3 | — | — |
| Cactus Needle 3 | 10 | 12 | 45% | 1 | — | — |
| Needle 26M | 6 | 3 | 67% | 1 | — | — |
What people said
40 of the 78 comments that mention it, newest first. Each links to the original comment.
“That looks like a really nice and well put together library.”
“I'm a big fan of Cactus's work on the needle family of Simple Attention Network”
“i made it run on esp32s3, and it works!!!”
“models of this size can't reason about relative values very well”
“they're incredibly cheap and handle NLP (not in a predictable way of course, but with great success rates)”
“to set the mode literally the opposite of what was requested.”
“Since it seems limited to matching a few templates and otherwise falling flat on its face”
“You explained it better, we've done a bad job at communicating its nice”
“Exactly the use cases we designed Needle for!”
“I don't think consumers would consider this "intelligence". It's too dumb.”
“it just completely ignored the "brightness" parameter”
“Tested with german and this kinda works, but confidence suffers.”
“if the number of cases where it works as expected is lower than what we could catch with a (old-Siri-style) heuristics based approach”
“I like Needle a lot, and I love the idea of any tiny resource-thrifty LLM that can run on older hardware, that outputs only JSON!”
“I'm quite impressed by the results of the web demo, especially given its size”
“It seems to me that the model struggles to have enough general intelligence, knowledge, or reasoning capacity”
“It fails a pretty simple navigation prompt.”
“It's a very small model, there's not much reasoning capability”
“being able to do that on a small embedded device is the novelty here”
“This is very interesting! I'm going to spend some time with this.”
“I added needle support for tool use in my side project.”
“limited to matching a few templates and otherwise falling flat on its face”
“It does better with clearer tool description”
“the web demo is not particularly impressive. It really doesn't like anything I throw at it.”
“It seems to convert simple instructions to reasonable tool calls.”
“Your `Fine-tuning` feature is even much more convenient.”
“it did a pretty good job for tool calling when I tried it”
“a file picker that renders in your current window (like oil) and uses more intelligent signals for ranking”
“any attempt to summarize parts of this discussion result in empty output: []”
“Which is kind of a little way there but not useable on its own”
“Context definitely helps. But yeah the quality of it doesn't seem to be too high.”
“Using agent for simple tasks is inefficient and wasteful, Needle really resolves this.”
“it's completely fascinating all on it's own”
“this could be a nice model to have as the first pass in a more complex agent system”
“This is a way that everyone can get that on the actual Apple Watch or local phone”
“This would be amazing for home assistant.”
“This is pretty much exactly what I want for Home Assistant.”
“This is very cool I'm going to try to carve out some time to try building this into my MOO system”
“Needle app or NotebookLM are also valuable resources for that.”
“Can answer anything about what happened in RAG in 2025.”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.