LFM3-1.7B-MLX-5bit Text Generation Model: 4GB+ VRAM

LFM3-1.7B-MLX-5bit Text Generation Model: 4GB+ VRAM

At a Glance

Item Value
Repository LiquidAI/LFM3-1.7B-MLX-5bit
Publisher guide Liquid AI: models and licenses
Published 2026-10-27
License lfm1.0
Formats MLX / safetensors
Source type Primary source (the publisher itself)

Values determined by this site’s code when the information was collected. Dates are JST.

Overview

LiquidAI has released “LFM3-1.7B-MLX-5bit", a 5-bit MLX quantization for Apple Silicon of their agent-focused model “LFM3-1.7B".

The base model “LFM3-1.7B" is a text generation model developed for agent capabilities and specialized applications running in local environments. It adopts a unique tri-hybrid architecture and can process a long-context window of up to 256K tokens.

Specifications

  • Parameters: 1.73B
  • Architecture: 24 layers: 13 gated short convolutions, 5 sliding-window attention, 6 global attention
  • Context Length: 256K tokens (input + output)

Performance

Below are the benchmark evaluation results listed in the model card of the source base model “LFM3-1.7B". These scores are actual measured values published by the release source using vLLM, and are figures from the unquantized model. To make the comparison easier to read, some model columns (Qwen3.5 9B) have been omitted from the model card table.

→ Scroll horizontally to see all columns

Benchmark LFM3 1.7B LFM2.5 2.6B MiniCPM5 2B Gemma 4 E4B Qwen3.5 4B
Agents
LiveClaw 45.60 14.70 30.37 20.36 19.94
Claw-Eval General (EN), average 68.64 62.63 60.74 56.14 61.96
Claw-Eval General (EN), pass^3 51.40 37.52 34.58 22.43 42.99
BrowseComp-Plus (OpenClaw subset) 37.00 26.33 16.83 21.00 24.67
WebArena-Verified 38.42 3.02 9.11 17.06 21.92
EnterpriseOps-Gym 28.20 7.94 13.41 9.86 25.96
AutomationBench (public) 41.29 0.94 4.71 1.92 8.67
Tool use
BFCL V4 71.39 55.89 62.65 48.25 55.12
ToolSandbox 80.36 78.04 68.16 69.27 75.40
MCP-Atlas 32.57 16.45 26.38 6.19 41.37
τ³-bench (4 domains), average 68.01 56.77 69.44 33.53 65.98
Agentic coding
SWE-bench Verified (OpenCode) 43.40 18.00 34.80 0.50 41.20
SWE-bench Multilingual (OpenCode) 23.00 8.00 15.67 0.67 23.33
Terminal-Bench 2.1 (OpenCode) 12.92 0.00 5.62 0.00 13.48
Coding
LiveCodeBench v6 73.46 60.57 70.52 71.00 61.99
LiveCodeBench Pro 25Q2 31.74 19.16 29.34 29.94 24.25
SciCode 28.47 16.32 27.78 25.00 23.02
Instruction following
IFEval 96.66 95.39 87.83 90.02 92.52
IFBench 73.03 60.48 62.08 44.56 38.16
IFStruct 93.72 82.18 72.65 78.45 87.05
Multi-IF 89.10 84.80 73.19 81.03 80.32
Knowledge
GPQA Diamond 63.37 57.78 65.15 57.88 75.76
Humanity’s Last Exam (no tools) 7.17 5.88 8.20 4.77 12.84
Humanity’s Last Exam (with tools) 13.90 9.60 11.50 6.00 11.60
AA-Omniscience (public) 0.50 -33.17 -5.00 -15.33 -57.17
AA-Omniscience accuracy (public) 7.07 8.75 9.13 9.83 15.67
AA-Omniscience non-hallucination (public) 92.93 53.69 84.44 72.09 13.64
Long context
AA-LCR 52.83 6.00 49.34 39.33 52.67
RULER 128K 65.56 1.15 37.96 18.82 84.63
Average 49.37 31.55 39.19 31.74 39.27

According to the comparison data in the model card, LFM3-1.7B shows particularly high scores in the fields of agent capabilities, tool use, and instruction following. It recorded 71.39 on BFCL V4, which measures function selection and argument assembly precision, the highest figure among the compared models. It also achieved an excellent score of 96.66 on IFEval, which measures compliance with formatting instructions. It also recorded 43.40 on SWE-bench Verified (OpenCode), which measures the ability to fix real GitHub issues, showing strengths in agentic coding tasks. On the other hand, it stays at 63.37 on GPQA Diamond, which handles PhD-level problems, falling short of Qwen3.5 4B’s 75.76, and is 65.56 on RULER 128K, which measures long-context effectiveness, resulting in scores that trail competing models in some knowledge and long-context tasks.

Strengths and Use Cases

The base model LFM3-1.7B is designed assuming local-environment agent capabilities, tool use, structured outputs, and domain-specific fine-tuning. It supports agent harnesses such as Hermes Agent, Pi, and OpenClaw, and is capable of processing utilizing a 256K token context window. Note that due to the small-scale characteristics of the model, it is not recommended for agentic coding or complex multi-step reasoning. Additionally, combining it with search or information retrieval is recommended for knowledge-intensive tasks.

Hardware Requirements

Estimated requirements (calculated by Local Model Watch) — 1.7B parameters (taken from the base model LiquidAI/LFM3-1.7B)

Your VRAM Quantization File size Est. memory needed
4GB (laptop iGPU / phone class) MLX 5bit 1.1GB 1.3GB

Inference engine support (architecture name matched against each project’s own model registry in its source code, checked 2026-11-17): llama.cpp: registered, vLLM: registered, MLX (mlx-lm): not registered. “Not registered" means the name is absent from that registry today, not that the model cannot run.

Memory estimates add a 20% runtime overhead (KV cache, etc.) to the actual size of the distributed files. Actual usage varies with context length, batch size and inference engine. These figures are computed by this site from file sizes, not published by the model’s authors. Compare with other models in our VRAM quick reference. What the quantization names mean: glossary.

Can You Run It Locally?

Not usable in Ollama, LM Studio and llama.cpp yet — we have found no GGUF build.

The publisher ships MLX / safetensors only. However, llama.cpp’s registry does list this architecture, so conversion to GGUF is possible and the model will run once someone publishes a converted build. Today it can be run with transformers, vLLM or MLX (Apple Silicon), using the memory figures in the table above.

License — lfm1.0 (Commercial use allowed with conditions): LiquidAI’s LFM Open License v1.0. Commercial use is allowed, except by legal entities with annual revenue of $10 million or more (Section 5). Redistribution requires the license text and notice of changes.

Compression: the MLX 5bit build measures 5.47 bits per weight — about 34% the size of the original 16-bit weights, calculated by this site from the actual file sizes.

Compiled by this site’s code from the published formats, converted builds we have found, and each engine’s own model registry. “Not found" means we have not seen such a build, not that none exists. License summaries are not legal advice — check the publisher’s original terms before relying on them.

Recent Models in the Same Size Class

Models with up to 4B parameters that Local Model Watch covered recently, listed by code from our article log for comparison. VRAM tiers are this site’s estimates; licenses are as stated on the model cards.

→ Scroll horizontally to see all columns

Model Parameters Smallest VRAM tier License Our article
openbmb/MiniCPM-V-4.7-35B-A3B-DSpark-gguf 389M 4GB apache-2.0 MiniCPM-V-4.7-35B-A3B-DSpark-gguf Draft Model for Speedup: 4GB+ VRAM (2026-11-17)
openbmb/MiniCPM-V-4.7-1B-DSpark-gguf 214M 4GB apache-2.0 MiniCPM-V-4.7-1B-DSpark-gguf Text Generation Model: 4GB+ VRAM (2026-11-17)
IFM/Jais-3-0.9B-GGUF 1.1B 4GB apache-2.0 Jais-3-0.9B-GGUF Bilingual Dense Model: 4GB+ VRAM (2026-11-17)
nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16 4.0B 4GB openmdw-1.1 NVIDIA Releases Dory1.0-4B: A Recurrent 4B LLM with Dynamic Loops (2026-11-17)
LiquidAI/LFM2.5-24B-A2B-DSpark 328M 4GB lfm1.0 LFM2.5-24B-A2B-DSpark Speculative Decoding Draft Model: 4GB+ VRAM (2026-11-17)

How to Get It

  • Distribution format: MLX, safetensors
  • Download command:
mlx_lm.generate --model LiquidAI/LFM3-1.7B-MLX-5bit \
  --prompt "What is the capital of Australia?" \
  --temp 0.1 --top-k 50 --max-tokens 16384
  • Supported engines: mlx-lm (support is reported to be in progress)

Related Articles

What to Read Next

Sources