LFM3-1.7B: VRAM requirements by quant
Key facts
Measured downloads are 1.10 GB Q4_K_M, 1.05 GB QAD-Q4_0, 1.43 GB Q6_K, 1.85 GB Q8_0, 3.47 GB BF16 and 0.97 GB MLX-4bit. LFM3-1.7B needs about 2.0 GB at standard quality (Q4, estimate).
Checked . LiquidAI Q4_K_M GGUF.
LFM3-1.7B is a small, fast AI model you can download and run on your own computer. It needs about 2 GB of memory at standard quality.
LIVE IN BROWSER · $0 · NO LOGIN
Test comparable open models (Recursive Flint 40B vs Gemma 4 31B) free in your browser
Open full 8-model playground ↗Click a starter prompt below or type your own question to watch Recursive Flint stream side by side against Gemma 4 31B right on this page:
Memory needed
About 2 GB
At standard quality, with room for a conversation of about 12 pages.
Get a 1-line alert when a faster quant for LFM3-1.7B drops — or when GPUs that run it drop >10%
Get a 1-line alert when a new open-weight model fits 8 GB VRAM (2.0 GB at Q4 for LFM3-1.7B) or when used RTX 3090 / 4090 & Mac Studio prices drop >10% — weekly, zero spam.
Machines that run LFM3-1.7B
Each of these holds the whole model in memory. Speed is in tokens per second (tok/s); from about 8 it keeps up with your reading.
EU shops · Shown to visitors in Europe.US shops · Shown to visitors outside Europe.
If you buy through these links we may earn a commission. It doesn't change our ranking. How shop links work
Mac mini M4 16GB
16 GB shared memory · second-hand
Fast~47 tok/s$615€549usedunverified
Machine details →Mac mini M6 16GB
16 GB shared memory
Fast~60 tok/s$989€1,069new
Machine details →Used PC with RTX 3090 24GB
24 GB graphics memory · second-hand
Fast~538 tok/s$2,413€2,346estimateused parts
Machine details →The buttons look up the graphics card, the part of this PC worth comparing.
PC with Radeon RX 7900 XTX 24GB
24 GB graphics memory · second-hand
Fast~532 tok/s$2,422€2,354estimateused parts
Machine details →The buttons look up the graphics card, the part of this PC worth comparing.
PC with RTX 5090 32GB
32 GB graphics memory
Fast~938 tok/s$7,834€8,467estimate
Machine details →The buttons look up the graphics card, the part of this PC worth comparing.
Where our price is for a used card or Mac, these shops mostly sell new ones, so their price can be higher.
As an Amazon Associate I earn from qualifying purchases.
Als Amazon-Partner verdiene ich an qualifizierten Verkäufen.
Als Amazon-Partner verdien ik aan in aanmerking komende aankopen.
About these prices
Each price is the amount we last wrote down for that machine, with its date on the machine page. "Estimate" is our own figure: for a complete PC we add our estimate for the other parts to the graphics card price, and some Mac prices are estimates, not a checked shop listing. "Used" is a second-hand price. "Unverified" means we have no checked shop listing for that exact amount. The shop buttons open a search for that machine at each shop, so you see today's price there. Prices excl. shipping unless marked. Euro prices incl. VAT where the seller charges it. Dollar prices are converted from our euro prices without the 21% EU VAT, before US sales tax.
Or rent one by the hour: RTX A6000 at Vast.ai holds it, for $0.40 an hour (median of live offers, 21 Oct 2026). Compare GPU cloud prices
Speeds and memory are our estimate from the hardware specs, not a measured test. Prices are dated and may change. How we calculate
Memory at each quality level
A lower quality level makes the model smaller, so it fits on a smaller machine. Standard (Q4) is what most people run.
| Quality | Memory needed |
|---|---|
| Q4 · standard (smallest that keeps quality) | 2.0 GB |
| Q5 · a bit sharper, a bit bigger | 2.1 GB |
| Q8 · near full quality, almost twice Q4 | 2.8 GB |
| FP16 · full quality, over three times Q4 | 4.4 GB |
With a conversation of 8,192 tokens · about 12 pages of text.
Sources and method
At standard quality: 1.1 GB for the model itself (a measured file), 0.1 GB for a conversation of 8,192 tokens, and 0.8 GB for the software.
Size source: huggingface.co/LiquidAI/LFM3-1.7B. The published size is rounded, so ours is an estimate.
Dense text model for local agents, 1.73B parameters. Hugging Face safetensors metadata lists 1,728,801,536 parameters, all BF16. 1.73B is that count to two decimals, so our parameter count is an estimate. The model card also states 1.73B. Architecture Lfm3ForCausalLM (model type lfm3): 24 layers, 13 gated short convolutions, 5 sliding-window attention layers with a 256-token window and 6 global attention layers, 4 KV heads, head dim 128, max context 262,144 (the card calls this 256K). Our estimate has only two attention buckets, so the 18 layers that are not global attention, including the 13 convolutions, are counted with that 256-token window. Conversation memory is therefore a little higher than the 6 global and 5 sliding layers alone. At 8,192 tokens that estimate is 0.11 GB of conversation memory plus 0.80 GB of working space. LiquidAI's Q4_K_M measured 1.1 GB on 28 Oct 2026 (1,099,627,808 bytes), so the total at 8,192 tokens is 2.01 GB. Q8_0 measured 1.85 GB (1,846,238,496 bytes) and BF16 measured 3.47 GB (3,466,626,336 bytes). F16 is the same 3,466,626,336 bytes. QAD Q4_0 measured 1.05 GB (1,049,615,680 bytes). Official MLX 4-bit weights measured 0.97 GB (973,083,386 bytes). Official GGUF, MLX and ONNX builds are published. ONNX is not in the fit table. LFM Open License v1.0 allows commercial use below $10 million annual revenue. The card says that at 4-bit precision the model takes about 1 GB and fits a phone or laptop. That is Liquid AI's statement. Our total adds conversation memory and working space to the measured file. The GGUF card says the files are for local inference on CPUs and GPUs, and that llama.cpp needs an LFM3-compatible build.
Licence
Commercial use: yes, with limits
Commercial use is allowed below $10 million annual revenue. At or above that, commercial use is not licensed.
Licence: LFM Open License v1.0 · checked 28 Oct 2026
Our reading of the licence, not legal advice.
Will LFM3-1.7B run on my graphics card?
Common questions
- How much memory does LFM3-1.7B need?
- About 2.0 GB at standard quality (Q4) with a conversation of 8,192 tokens: 1.1 GB for the model itself, 0.1 GB for the conversation and 0.8 GB for the software. The model size is a measured file; the conversation and software parts are our estimate.
- Does LFM3-1.7B fit on a single 24 GB graphics card?
- Yes, by our estimate. It needs about 2.0 GB, and we count 23.4 GB of a 24 GB card as usable.
- Are the speeds measured?
- No. They are our estimate from how fast each machine's memory is. Real speed also depends on the software and its settings.
- How much memory does LFM3-1.7B need at Q4_K_M, Q8_0 and BF16, and does it fit a 4 GB GPU, laptop or phone?
- At Q4_K_M (1.1 GB measured GGUF + 0.11 GB conversation memory at 8,192 tokens + 0.80 GB working space = 2.01 GB / 2.0 GB total estimate). At Q8_0 (1.85 GB measured GGUF + 0.11 GB conversation memory at 8,192 tokens + 0.80 GB working space = 2.76 GB / 2.8 GB total estimate). At BF16 (3.47 GB measured GGUF + 0.11 GB conversation memory at 8,192 tokens + 0.80 GB working space = 4.38 GB / 4.4 GB total estimate). The QAD Q4_0 file is QAD-Q4_0 (1.05 GB measured GGUF, 2.0 GB total estimate). A 4 GB card would leave 3.4 GB usable under the same 0.6 GB reserve this site uses on a graphics card (this list has no 4 GB card). Q4_K_M fits that 3.4 GB. BF16 at 4.4 GB does not fit that 3.4 GB. GeForce RTX 3060 12GB (11.4 GB usable) holds the Q4_K_M file, and so does every larger card on https://localllmrig.com/can-i-run/lfm3-1.7b. The model card says that at 4-bit precision it takes about 1 GB of memory and fits a phone or laptop. That is Liquid AI's statement. Our total adds conversation memory and working space to the measured file. MLX-4bit (0.97 GB measured MLX, 1.9 GB total estimate) is the official Apple Silicon build.
- Can I run LFM3-1.7B on a MacBook Air or Mac mini M4, and should I use MLX or GGUF?
- Yes. Liquid AI publishes MLX weights for Apple Silicon and GGUF files for llama.cpp, Ollama and LM Studio. MLX-4bit is 0.97 GB measured (1.88 GB total estimate at 8,192 tokens). MLX-bf16 is 3.46 GB measured (4.37 GB total estimate). An 8 GB Mac leaves 5.3 GB usable under the same two-thirds graphics share this site uses for Macs up to 32 GB, so both of those totals fit. Mac mini M4 16GB leaves 10.7 GB usable of 16 GB (about €549 ($615), about 47 tok/s estimate at Q4). Use MLX on Apple Silicon. Use GGUF when the runtime is llama.cpp, Ollama or LM Studio. The GGUF card says llama.cpp needs an LFM3-compatible build.
- How does LFM3-1.7B compare with Spark-X2.6-4B and NVIDIA Nemotron 3.5 Bolt E4B on 8 GB hardware?
- All three fit an 8 GB GPU (7.4 GB usable) at their standard 4-bit file. LFM3-1.7B (1.73B parameters, our estimate) needs 2.01 GB total at Q4_K_M (1.10 GB measured GGUF) and 4.38 GB at BF16, so BF16 also fits that 7.4 GB. Spark-X2.6-4B (4.11B parameters, our estimate) needs 3.78 GB total at bartowski Q4_K_M (2.62 GB measured GGUF) and 4.70 GB for the official 3.54 GB Q4_K_M file. Its Q8_0 total is 5.54 GB. Its BF16 file is 8.23 GB (9.39 GB total), which fits a 12 GB GPU (11.4 GB usable). On GeForce RTX 3060 12GB the Q4 speed estimate is about 93 tok/s. NVIDIA Nemotron 3.5 Bolt E4B is 8B parameters. The card calls 4B of that effective, from per-layer embeddings. Its measured Q4_K_M file is 6.18 GB (7.25 GB total at 8,192 tokens), which fits the 7.4 GB an 8 GB GPU leaves usable. Its BF16 file is 15.05 GB (16.12 GB total), which fits a 24 GB GPU (23.4 GB usable). The context stored for it is 65,536 tokens. The GGUF header and Ollama list 128K.
- Why is LFM3-1.7B's conversation memory 0.11 GB at 8,192 tokens when the max context is 262,144?
- LFM3-1.7B uses Lfm3ForCausalLM (lfm3): 24 layers, 13 gated short convolutions, 5 sliding-window attention layers with a 256-token window, and 6 global attention layers (4 KV heads, head dim 128). Our estimate has two buckets, so the 18 layers that are not global attention, including the 13 convolutions, are counted with that 256-token window. Conversation memory is therefore a little higher than the 6 global and 5 sliding layers alone. At 8,192 tokens that estimate is 0.11 GB. At the full 262,144-token context it is 3.23 GB.
- Does Lfm3ForCausalLM run in llama.cpp, Ollama, LM Studio, MLX, ONNX, vLLM or Transformers?
- The LFM3-1.7B model card lists GGUF for llama.cpp, Ollama and LM Studio, MLX for Apple Silicon, ONNX for ONNX Runtime, and BF16 weights for vLLM, SGLang and Transformers. The GGUF card says llama.cpp needs an LFM3-compatible build, and that the GGUF files are for local inference on CPUs and GPUs. The Transformers example on the model card sets trust_remote_code because the repository ships custom code. ONNX is published and is not in the fit table on https://localllmrig.com/can-i-run/lfm3-1.7b.
- Where can I test open models in the browser before downloading LFM3-1.7B?
- LFM3-1.7B is not one of the 10 free models. You can test those open models in the browser at https://localllmrig.com/ with no login and no local graphics memory, then read the fit on https://localllmrig.com/can-i-run/lfm3-1.7b.
- What licence does LFM3-1.7B use, and is commercial use allowed?
- LFM3-1.7B is listed as LFM Open License v1.0. Our reading of the licence is that commercial use is allowed below $10 million annual revenue, and is not licensed at or above that. That is our reading of the licence, not legal advice. File sizes and the machines that hold them are on https://localllmrig.com/can-i-run/lfm3-1.7b. You can try open models live for free at https://localllmrig.com/.