LFM2.5 2.6B
Ranked 6 of 11 small language models by recommenders
Fast small model people run locally for classification, security triage, and simple tool use on modest hardware. Critics report coherence problems and say it is weaker than similar-sized rivals.
Fast small model people run locally for classification, security triage, and simple tool use on modest hardware. Critics report coherence problems and say it is weaker than similar-sized rivals.
22 people gave an opinion
- recommend it
- 14
- criticize it
- 8
- approval share
- 64%
Recommended in 9 of 13 months
- November 2025: not recommended
- December 2025: not recommended
- January 2026: recommended
- February 2026: recommended
- March 2026: recommended
- April 2026: recommended
- May 2026: recommended
- June 2026: recommended
- July 2026: not recommended
- August 2026: recommended
- September 2026: recommended
- October 2026: recommended
- November 2026: not recommended (month in progress)
Why it doesn’t carry the badge
- 14 people recommend itNeeds 10 or more, met.
- 1.7 recommenders for every criticNeeds 2 to 1 or better, not met.
- Recommended in 9 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
Fast on modest or older hardware
“I love these smaller models for the added capability they can bring to older hardware.”
Comment from 11 Aug 2026 in “LFM2.5 2.6B model competitive with 4x larger models” “they were so fast that you could use them in a similar way”
Comment from 7 Oct 2026 in “Strands Decider 2B: a small, open-source, decision model” “With good speed even on a 12th gen Intel.”
Comment from 11 Aug 2026 in “LFM2.5 2.6B model competitive with 4x larger models” Good at classification and security tasks
“You can recreate most of those with LFM 2.5 classifier locally”
Comment from 19 Sep 2026 in “OpenJev” “you can just prompt for classification”
Comment from 27 Oct 2026 in “Introducing LFM3: Every Modality for Specialized Intelligence” “handle their security job nicely”
Comment from 20 Mar 2026 in “MacBook M5 Pro and Qwen3.5 = Local AI Security System” Vision variant for vision tasks
“LiquidAI's LFM 2.5 uses novel techniques to eliminate doom loops, and they have a vision variant”
Comment from 11 May 2026 in “Local AI needs to be the norm” “My prefer is LFM 450M for vision task, QWEN 9B Q4 for Orchestration”
Comment from 20 Mar 2026 in “MacBook M5 Pro and Qwen3.5 = Local AI Security System” Pairs with larger model for orchestration
“My prefer is LFM 450M for vision task, QWEN 9B Q4 for Orchestration”
Comment from 20 Mar 2026 in “MacBook M5 Pro and Qwen3.5 = Local AI Security System” “So I like LFM 450M + QWEN 9B Q4”
Comment from 20 Mar 2026 in “MacBook M5 Pro and Qwen3.5 = Local AI Security System”
What people warn about
Each mark is one comment
Serious coherence issues
“They have serious issues with coherence.”
Comment from 11 Aug 2026 in “LFM2.5 2.6B model competitive with 4x larger models” “suffering from serious coherence issues”
Comment from 2 May 2026 in “LFM2-24B-A2B: Scaling Up the LFM2 Architecture” Not competitive with similar-sized models
“I don't think it's even competitive with Gemma 4 E2B”
Comment from 11 Aug 2026 in “LFM2.5 2.6B model competitive with 4x larger models” “It's not even competitive with 2x sized Qwen 4B.”
Comment from 11 Aug 2026 in “LFM2.5 2.6B model competitive with 4x larger models” Often underperforms in practice
“These LiquidAI models have never worked well for me in practice.”
Comment from 11 Aug 2026 in “LFM2.5 2.6B model competitive with 4x larger models” “2.5 version was too dumb for HA”
Comment from 27 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 | Criticize | Approval | Months | Owned | Badge |
|---|---|---|---|---|---|---|
| Claude Haiku | 218 | 134 | 62% | 13 | 4 mo4 owners | — |
| OpenAI Luna | 218 | 56 | 80% | 7 | — | Approved |
| Apple Foundation Models | 45 | 35 | 56% | 11 | 1 yr4 owners | — |
| Ministral | 18 | 6 | 75% | 8 | — | Approved |
| LFM2.5 2.6BThis page | 14 | 8 | 64% | 9 | — | — |
What people said
27 of the 29 comments that mention it, newest first. Each links to the original comment.
“you can just prompt for classification”
“2.5 version was too dumb for HA”
“they were so fast that you could use them in a similar way”
“You can recreate most of those with LFM 2.5 classifier locally”
“Liquid has released 230M and 350M models that work far, far better in my testing”
“LFM2.5-2.6B overtook both”
“I love these smaller models for the added capability they can bring to older hardware.”
“I don't think it's even competitive with Gemma 4 E2B”
“it can still be quite good for question answering and such”
“they didn't want to call tools in the same way as other models, and it led to bare tool calls in the response”
“They have serious issues with coherence.”
“It's not even competitive with 2x sized Qwen 4B.”
“With good speed even on a 12th gen Intel.”
“There's a strong incentive to cherry pick in self-reported comparisons.”
“We recommend using it for agentic workloads, tool use, data extraction, RAG, and long-context workflows.”
“If you're looking for good performance out of tiny models, LFM has the most advanced design.”
“These LiquidAI models have never worked well for me in practice.”
“LiquidAI LFM models are amazing, but very situational.”
“it's good at instruction following as long as the instruction is short and simple”
“LiquidAI's LFM 2.5 uses novel techniques to eliminate doom loops, and they have a vision variant”
“suffering from serious coherence issues”
“taking a small language model (LiquidAI's LFM2-2.6B) and turning it into a Tic-Tac-Toe master that beats GPT-5-mini”
“handle their security job nicely”
“So I like LFM 450M + QWEN 9B Q4”
“this provides a good result”
“My prefer is LFM 450M for vision task, QWEN 9B Q4 for Orchestration”
“for edge running look into the LFM 2.5 class of models”
Counts come from 1,795,640 Hacker News comments, November 2025 to November 2026. How we count. Report a mistake.