LFM3
Ranked 9 of 10 small language models by recommenders
People recommend the small LFM3 models for strong text ability relative to their size and for running on modest memory. Critics say its coding output is weak and its answers are sometimes inaccurate.
7 people recommended LFM3 in Hacker News comments, 3 had mixed feelings and 2 criticized it, November 2025 to October 2026.
7 of the 7 people who recommend it commented in one thread, “Introducing LFM3: Every Modality for Specialized Intelligence”.
Summary written by Claude from 15 comments.
12 people gave an opinion
- recommend it
- 7
- mixed feelings
- 3
- warn against or regret it
- 2
The record
Recommended in 1 of 12 months
In 1 different thread.
- 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: not recommended
- October 2026: recommended
Why it doesn’t carry the badge
- 7 people recommend itNeeds 10 or more, not met.
- 3.5 recommenders for every criticNeeds 2 to 1 or better, met.
- Recommended in 1 of 12 monthsNeeds 4 or more, not met.
- Recommended in 1 threadNeeds 5 or more, not met.
- 7 of its 7 recommenders came from one threadNeeds half or less, not met.
The same rules apply to every product, and the badge can’t be bought. Mixed feelings count toward neither side. How the badge works
What people say about it
Reasons grouped from a sample of 15 of the 18 opinions about it, picked in proportion to how many people recommend it, have mixed feelings and criticize it. Each mark is one comment in the sample. Open a reason to read what they said.
Why people recommend it
Each mark is one comment
Strong for its size
“It's a model designed to be tiny and still functional, unlike other larger models that are just scaled down”
Comment from 27 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence” “LFM3-1.7B is better at pure text tasks than all other models”
Comment from 28 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence” “I think the d1-2B is amazing, it can run on 1GB of unified RAM.”
Comment from 27 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence” - Comment from 27 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence”
Step up from earlier versions
“Excited to roll out LFM 3 to replace it, looks like a big step up.”
Comment from 27 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence” “to beat Qwen3.5-4B with such a capable model is amazing”
Comment from 27 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence” Runs in about 1GB memory
“LFM3-1.7B fits in about 1 GB at 4-bit, holds 256K context”
Comment from 27 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence”
What people warn about
Each mark is one comment
Weak at coding, spaghetti code
“definitely would not recommend using it for SWE-style tasks. But chaining a couple bash calls together to find a file? no problem”
Comment from 28 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence” “It will produce code that mixes too many things at once, creating spaghetti code”
Comment from 29 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence”
Other small language models people recommend
The four most recommended other small language 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 | Mixed | Criticize | Approval | Months | Had for | Badge |
|---|---|---|---|---|---|---|---|
| OpenAI Luna | 171 | 31 | 46 | 79% | 4 | — | Approved |
| Claude Haiku | 147 | 48 | 113 | 57% | 12 | — | — |
| Apple Foundation Models | 24 | 17 | 31 | 44% | 10 | — | — |
| Ministral | 14 | 3 | 5 | 74% | 7 | — | Approved |
| LFM3This page | 7 | 3 | 2 | 78% | 1 | — | — |
Also compared with LFM2.5 2.6B.
What people said
15 of the 18 opinions about it, picked in proportion to the three counts, newest first. Each links to the original comment.
“It will produce code that mixes too many things at once, creating spaghetti code”
“some initial testing shows useful-but-sometimes-inaccurate”
“definitely would not recommend using it for SWE-style tasks. But chaining a couple bash calls together to find a file? no problem”
“LFM3-1.7B is better at pure text tasks than all other models”
“First class hardware support makes this really exciting and different.”
“to beat Qwen3.5-4B with such a capable model is amazing”
“then LFM3-1.7B for extracting particular action items”
“LFM3-1.7B said that a cow can say “Oink”
“It's a model designed to be tiny and still functional, unlike other larger models that are just scaled down”
“none of them could actually fire the model up, kept saying safetensors cant be read”
“I’ve been using 2.6B as my main home assistant model and it’s been good.”
“Most impressive for a 1.7B model!”
“LFM3-1.7B fits in about 1 GB at 4-bit, holds 256K context”
“Excited to roll out LFM 3 to replace it, looks like a big step up.”
“I think the d1-2B is amazing, it can run on 1GB of unified RAM.”
Counts come from 1,791,911 Hacker News comments labeled by Claude, November 2025 to October 2026, with data through October 31, 2026. Counts updated 2 Nov 2026. How we count. Report a mistake.