# LFM2.5-24B-A2B by Liquid AI

23 Hacker News commenters recommended LFM2.5-24B-A2B by Liquid AI and 10 criticized it, November 2025 to November 2026. It was recommended in 2 of 13 months.

**Verdict:** Fast, efficient open-weight model for on-device agents; critics say it lags Qwen on knowledge and coding, and the license is confusing.

## Why people recommend it

- Fast output on modest hardware (5 comments): "fast is fast (18tk/s on a Intel Core 7 155H laptop), but quality is something else" ([source](https://news.ycombinator.com/item?id=50099399)) "I am getting at most 25tps with the llama cpp vulkan backend" ([source](https://news.ycombinator.com/item?id=47986534)) "this thing puts out results very fast even on max reasoning" ([source](https://news.ycombinator.com/item?id=50097887))
- Good for on-device work (3 comments): "Really Cool, the on-device shifts the economics a bit." ([source](https://news.ycombinator.com/item?id=50111471)) "It is a good model for on-device work." ([source](https://news.ycombinator.com/item?id=50099184)) "I just tried the excellent local model" ([source](https://news.ycombinator.com/item?id=50102801))
- Strong results for its size (3 comments): "Way more impressive than I expected. I am still messing about with it, but I think it's a keeper for general discussions." ([source](https://news.ycombinator.com/item?id=50246838)) "Wow, those are seriously impressive numbers for how intelligent this model seems to be!" ([source](https://news.ycombinator.com/item?id=50097492)) "someone has caught up to Qwen (albeit Qwen from 1-2 months ago) in parameter efficiency" ([source](https://news.ycombinator.com/item?id=50097853))
- Handles terminal and scripting tasks (3 comments): "it's stronger for terminal use, like chaining bash calls, monitoring background tasks, etc." ([source](https://news.ycombinator.com/item?id=50098760)) "these small agentic-focused models are a lot of fun to play with" ([source](https://news.ycombinator.com/item?id=50097716)) "its doing pretty well writing some python scripts for me" ([source](https://news.ycombinator.com/item?id=50242394))

## What people warn about

- Weak at coding and agentic loops (3 comments): "Even Liquid's own documentation states that coding isn't this model's strong suit." ([source](https://news.ycombinator.com/item?id=50098879)) "kept going in circles fixing one compile error and introducing another" ([source](https://news.ycombinator.com/item?id=50098482)) "So this wouldn't work in its current state in an agentic loop." ([source](https://news.ycombinator.com/item?id=50100012))
- Behind Qwen on benchmarks (2 comments): "it performs much, much worse than Qwen 35B A3B for general knowledge" ([source](https://news.ycombinator.com/item?id=50098818)) "it appears to get a worse score on SWE-Bench-Verified (by a lot) when compared to Qwen 3.6 35B-A3B" ([source](https://news.ycombinator.com/item?id=50097900))
- Unclear non-commercial license clause (2 comments): "this model has a questionably-enforceable non-commercial use clause in its license" ([source](https://news.ycombinator.com/item?id=50098829)) "The license doesn't really make much sense to me" ([source](https://news.ycombinator.com/item?id=50099639))
- Fast but lower quality output (3 comments): "fast is fast (18tk/s on a Intel Core 7 155H laptop), but quality is something else" ([source](https://news.ycombinator.com/item?id=50099399)) "this is fast but maybe not usable" ([source](https://news.ycombinator.com/item?id=50118560)) "It is very fast on a 4090, but also very stupid." ([source](https://news.ycombinator.com/item?id=50101497))

## Alternatives hackers recommend

- [Qwen 3.8 27B](https://hackerapproved.com/p/qwen-3-8-27b.md): 1173 recommend
- [Z.ai GLM](https://hackerapproved.com/p/glm-5-2.md): 1043 recommend
- [Kimi](https://hackerapproved.com/p/kimi-k3.md): 931 recommend
- [Gemma 4](https://hackerapproved.com/p/gemma-4.md): 713 recommend
- [Qwen](https://hackerapproved.com/p/qwen.md): 474 recommend

## Cite as

Hacker Approved, "LFM2.5-24B-A2B by Liquid AI", https://hackerapproved.com/p/lfm2-5-24b-a2b/ (counts from 1,795,640 Hacker News comments, November 2025 to November 2026).