LOADING DATASHEET
GROUNDED IN REAL HUMAN CONVERSATIONSYOUR BENCH
LFM2.5 2.6B
◆ OPEN
by Liquid
OPEN WEIGHTSLOCALFREE TO TRYCODING HELP
RANKED #170 OF 316 ASSISTANTS · OVERALL #233 OF 493 RANKED · VIBE SCORE 6.4 · 274 VOICES ·
COPY BADGE pastes a README snippet with the rank SVG.
A fast and free AI model that runs on your computer, perfect for quick coding help.
HAPPY9.9
99 mentions
SAD5.1
1 mentions
VIBE TRAJECTORY
26 weeks · 286 voices in this window
HOLDING UP
Recent vibe is steady. No clear fade in the crowd.
HOW TO USE IT
Open weights. You can use a hosted service, or download it and run it yourself, free.
RUN IT ON YOUR OWN MACHINE
Open weights: download LFM2.5 2.6B once and it's yours. No subscription, no rate limits, works offline.
2.69B parameters, a Q4 file is about 1.7 GB
- BARE MINIMUM
Raspberry Pi 5 (8GB RAM)
8 GB RAM
~$80 NEWQ4_K_M, ~10 tok/s, 4k context
- SWEET SPOT
Apple Mac Mini M1 (8GB RAM)
8 GB UNIFIED MEMORY
~$300 USEDQ8_0, ~90 tok/s, 32k context
- FULL POWER
NVIDIA GeForce RTX 3060 (12GB VRAM)
12 GB VRAM
~$200 USEDBF16 (unquantized), ~150+ tok/s, full 128k context
LEARN HOW TO SET IT UP
WILL MY BOX RUN LFM2.5 2.6B?
The three cards above are researched picks. Search the machine you actually have — we only say yes if it should reply at a usable speed, not just load the weights and crawl.
HARDWARE PICKS & PRICES ARE RESEARCHED FROM THE LIVE WEB AND REFRESHED AUTOMATICALLY. TREAT THEM AS BALLPARK, NOT GOSPEL.
WHAT CAN YOU ACTUALLY DO WITH IT?
Everyday jobs on each of those rigs: how long LFM2.5 2.6B takes, and how good the crowd says it is at that kind of work.
| THE JOB | BARE MINIMUMRaspberry Pi 5 (8GB RAM) | SWEET SPOTApple Mac Mini M1 (8GB RAM) | FULL POWERNVIDIA GeForce RTX 3060 (12GB VRAM) |
|---|---|---|---|
| Write an emaila paragraph or two, drafted from a one-line brief | ~50 S | ~6 S | NO DATA |
| Summarize a documenta long report boiled down to the points that matter | ~2.5 MIN★★★★★ | ~17 S★★★★★ | —★★★★★ |
| Build a websitea small landing page, markup and styles together | ~13 MIN★★★★★ | ~89 S★★★★★ | —★★★★★ |
| Build a backendan API with routes, storage and tests | ~42 MIN★★★★★ | ~4.6 MIN★★★★★ | —★★★★★ |
| Build a gamea playable browser game in one file | ~25 MIN★★★★★ | ~2.8 MIN★★★★★ | —★★★★★ |
STARS ARE THE COMMUNITY SCORE FOR THAT KIND OF WORK, DOCKED FOR HOW SQUASHED THE WEIGHTS GET AT EACH TIER. A DASH MEANS NO SPEED WAS RESEARCHED FOR THAT RIG. TIMES ARE COMPUTED FROM RESEARCHED SPEEDS. BALLPARK, NOT GOSPEL.
WHAT THE INTERNET SAYS
People find LFM2.5.2.6B to be a fast, dependable, and highly valuable tool, especially for coding help. However, some users are frustrated by its tendency to lecture or refuse requests.
259 POSTS · 19 COMMENTS · HACKER NEWS · X · GITHUB · REDDIT · DEV.TO · BLUESKY · LEMMY · BLOG
PEOPLE LOVE
- Stable and dependable (2 mentions)
PEOPLE COMPLAIN ABOUT
- Outages, errors, or limits get in the way (1 mentions)
TOPIC BY TOPIC
- Working when you need itMOSTLY POSITIVE
2 thumbs up · 1 thumbs down
LOOKING FOR AN ALTERNATIVE?
Checked picks first: tags say why you would switch, stars say how fully each one stands in for LFM2.5 2.6B.
Phi 4 Mini
Muse Spark 1.3
DeepSeek V4 Flash> SOURCES (12)
- REDDIT · 2026-08-04 · 220 UPVOTESA 2.6B model with tool calling and 128K context now runs at 30 tok/s on a phone
A 2.6B model with tool calling and 128K context now runs at 30 tok/s on a phone
- REDDIT · 2026-08-04 · 135 UPVOTESLFM2.5-2.6B is out
LFM2.5-2.6B is out
- REDDIT · 2026-08-07 · 103 UPVOTESLFM2.5-2.6B model+KV cache quantization report
LFM2.5-2.6B model+KV cache quantization report
- REDDIT · 2026-07-07 · 61 UPVOTESLiquid AI - Antidoom (the doom loop remover)
Liquid AI - Antidoom (the doom loop remover)
- REDDIT · 2026-07-02 · 41 UPVOTESMade a new 350M model to compete with lfm2.5 but with an open license
Made a new 350M model to compete with lfm2.5 but with an open license
- BLOG · 2026-08-05 · 32 UPVOTES[AINews] Megakernels are so dead and so back
Part of our Inference Engineering Masterclass pod yesterday involved a spicy discussion about Megakernels: megakernels are dead why are megakernels useful? you spend two months writing a kernel to save time on launch overhead and poor inter-kernel overlap.…
- X · 2026-08-16 · 9 UPVOTESAugust '26 LLM list (mid-month) - crazy big models this month Fable 5.1 ❌ OpenAI Astra ❌ Meta Watermelon ❌ Muse Spark 1.
August '26 LLM list (mid-month) - crazy big models this month Fable 5.1 ❌ OpenAI Astra ❌ Meta Watermelon ❌ Muse Spark 1.
- REDDIT · 2026-08-05 · 8 UPVOTESTested LFM2.5 2.6B on Agentic Work (Tool Calling) & Coding with OpenCode
Tested LFM2.5 2.6B on Agentic Work (Tool Calling) & Coding with OpenCode
- REDDIT · 2026-08-20 · 7 UPVOTESUp to 3.2x Faster Inference with LFM2.5-DSpark
Up to 3.2x Faster Inference with LFM2.5-DSpark
- X · 2026-08-14 · 4 UPVOTESDaily AI Briefing — Thursday, August 14, 2026 Executive Summary The past 48 hours were dominated by a speed and economic
Daily AI Briefing — Thursday, August 14, 2026 Executive Summary The past 48 hours were dominated by a speed and economic
- GITHUB · 2026-08-13 · 0 UPVOTESfeat(catalog): add LiquidAI LFM2.5-VL 3B GGUF (llama.cpp VLM)
feat(catalog): add LiquidAI LFM2.5-VL 3B GGUF (llama.cpp VLM)
- GITHUB · 2026-09-30 · 0 UPVOTESchore: pin llama.rn 0.13.0-rc.6 (with unload fix), cap low-RAM auto-pick at 1.1 GB, honest full-answer hint
chore: pin llama.rn 0.13.0-rc.6 (with unload fix), cap low-RAM auto-pick at 1.1 GB, honest full-answer hint