A model is still running

A model is still running or downloading. Leaving will stop the current operation. Stay on this page to let it finish.

Skip to content
LocalAI.toolsBeta

Join the waitlist

Get an email when the new apps and model runs launch.

No account needed.

Liquid LFM3 Desktop

Local text chat with Liquid LFM3 1.7B.

Desktop inferencePrivate · Runs on your computer
About this model

Answer written or spoken questions, transcribe WAV clips, or read short text aloud with Liquid AI LFM3 Audio.

Created by
Liquid AI
Model released
Unknown
Source repository created
Sep 8, 2026
Model weights · Q4
1.2 GB

Best for

  • Answer written questions, transcribe WAV clips, and synthesize short speech.
  • Answer short spoken questions using batch audio chat.

Keep in mind

  • Check transcription and factual answers against the source. Noise and accents can introduce errors.
  • The candidate accepts one WAV data URL per request. Live microphone sessions are outside this candidate interface.
  • The Liquid AI license places conditions on commercial use and redistribution.
  • Live ASR processes an English WAV in frames. Realtime conversational output failed a spoken instruction case. Other languages and audio tool calling have no qualification evidence.

Processed on your deviceYour input is processed locally in the desktop app. Model and runtime files download from external hosts.

Target platforms · verification required

  • macos · aarch64, x86_64
  • windows · x86_64

Recommended configuration

Q4 · metal (verified platform only)

Sources & technical details
Original model
https://huggingface.co/LiquidAI/LFM3-Audio-2B
Release verification
Unknown
Release evidence
Exact checkpoint date pending provenance review.
Source updated
Nov 3, 2026
Native model files
LiquidAI/LFM3-Audio-2B-GGUF
Files updated
Nov 3, 2026
Artifact revision
39670cb5c42e9c4bfe046519ba7ffbe47f9552ad

Desktop app

Sample result

LFM3 Audio 2B Desktop

Run Liquid LFM3 Desktop on your computer

Download the desktop app to load this model, provide input and view results locally.

Public desktop downloads are coming soon.

Device requirements

macOS Apple SiliconCPU / Q4TestedNot establishedRAM estimate
macOS Apple SiliconMETAL / Q4TestedNot establishedRAM estimate
If your device falls short

MemoryShow advisory requirements.

GPUUse verified CPU recovery when permitted.

PlatformNo verified configuration.

Benchmark evidence

Saved desktop measurements apply to the recorded hardware and runtime. Fallback cases report the device that actually ran; they do not prove GPU support on the requested device.

LFM3 Audio 2B Desktop

14 configurations tested
Passed14
Failed0
TestedNov 15, 2026
Runtimellama.cpp b12359

Quality criteria

No explicit quality gate recorded. Review per-case evidence; full certification is pending.

PrecisionDeviceResultPlatformInference / load
Q4metalAcceleratedPassedFastmacos · aarch64195 ms / 757 ms
Q4metalAcceleratedPassedFastmacos · aarch64220 ms / 655 ms
Q4metalAcceleratedPassedFastmacos · aarch64389 ms / 655 ms
Q4metalAcceleratedPassedFastmacos · aarch64586 ms / 655 ms
Q4metalAcceleratedPassedFastmacos · aarch64829 ms / 762 ms
Q4metalAcceleratedPassedFastmacos · aarch641097 ms / 655 ms
Q4metalAcceleratedPassedFastmacos · aarch643421 ms / 549 ms
Q4cpuPassedFastmacos · aarch64325 ms / 8097 ms
Q4cpuPassedFastmacos · aarch64677 ms / 2393 ms
Q4cpuPassedFastmacos · aarch64966 ms / 7333 ms
Q4cpuPassedFastmacos · aarch64975 ms / 7333 ms
Q4cpuPassedFastmacos · aarch64983 ms / 7333 ms
Q4cpuPassedFastmacos · aarch641099 ms / 6849 ms
Q4cpuPassedFastmacos · aarch641119 ms / 8682 ms
Test setup & sample details
Quality gate
No explicit quality gate recorded. Review the per-case quality evidence; full certification is pending.
Tested
2026-11-15T22:07:40.885Z
Environment
{ "scope": "DesktopAI TypeScript client using native JSON-lines transport with verified cached sources.", "hardware": "Apple M3 Max", "os": "macOS", "arch": "arm64", "runtime": { "name": "llama.cpp", "version": "b12359" }, "speedGradePolicy": "Short requests completing in 2 seconds or less are interactive." }
Runtime
llama.cpp b12359
Artifact revision
39670cb5c42e9c4bfe046519ba7ffbe47f9552ad
Fixture
{
  "file": "model-samples/voxtral-realtime/jfk.wav",
  "size": 352078,
  "sha1": "069db705beb788608e93b99f1a4300451ebfb972",
  "sha256": "4d968ac99a1d0d4bc42ae8dd1552f4235e7b952a8121b39797f3dc2ca16123e9",
  "url": "https://assets.localai.tools/samples/v1/files/4d968ac99a1d0d4bc42ae8dd1552f4235e7b952a8121b39797f3dc2ca16123e9/jfk.wav"
}
Rerun when
artifactRevision; runtimeVersion; fixtureChecksum; qualityGate; nativeBackend
Per-case quality & failure details
[
  {
    "precision": "q4",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "metric": "frozenAnswerMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "cpu",
      "cpuThreads": 4,
      "device": "cpu",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 974.928166,
      "loadMs": 7333.041291,
      "offloadedLayers": 0,
      "precision": "q4",
      "provider": "llama.cpp CPU",
      "requestedDevice": "cpu",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 8307.969457,
      "performanceMetric": {
        "name": "charactersPerSecond",
        "value": 2.0514331924635356
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "metric": "normalizedWordMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "cpu",
      "cpuThreads": 4,
      "device": "cpu",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 965.925417,
      "loadMs": 7333.041291,
      "offloadedLayers": 0,
      "precision": "q4",
      "provider": "llama.cpp CPU",
      "requestedDevice": "cpu",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 8298.966708,
      "performanceMetric": {
        "name": "audioSecondsPerSecond",
        "value": 11.38804280993467
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "metric": "independentWhisperWordMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "cpu",
      "cpuThreads": 4,
      "device": "cpu",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 982.7675409999999,
      "loadMs": 7333.041291,
      "offloadedLayers": 0,
      "precision": "q4",
      "provider": "llama.cpp CPU",
      "requestedDevice": "cpu",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 8315.808831999999,
      "performanceMetric": {
        "name": "audioSecondsPerSecond",
        "value": 2.7676941764196914
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "metric": "frozenAnswerMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "cpu",
      "cpuThreads": 4,
      "device": "cpu",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 677.300208,
      "loadMs": 2393.01375,
      "offloadedLayers": 0,
      "precision": "q4",
      "provider": "llama.cpp CPU",
      "requestedDevice": "cpu",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 3070.313958,
      "performanceMetric": {
        "name": "charactersPerSecond",
        "value": 13.288051433759488
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "metal",
    "status": "pass",
    "quality": {
      "metric": "frozenAnswerMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "partial",
      "cpuThreads": 4,
      "device": "metal",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 1097.195542,
      "loadMs": 655.4910000000001,
      "offloadedLayers": 25,
      "precision": "q4",
      "provider": "llama.cpp Metal",
      "requestedDevice": "metal",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 1752.686542,
      "performanceMetric": {
        "name": "charactersPerSecond",
        "value": 1.8228291343166978
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "metal",
    "status": "pass",
    "quality": {
      "metric": "normalizedWordMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "partial",
      "cpuThreads": 4,
      "device": "metal",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 388.617917,
      "loadMs": 655.4910000000001,
      "offloadedLayers": 25,
      "precision": "q4",
      "provider": "llama.cpp Metal",
      "requestedDevice": "metal",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 1044.108917,
      "performanceMetric": {
        "name": "audioSecondsPerSecond",
        "value": 28.30543708565038
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "metal",
    "status": "pass",
    "quality": {
      "metric": "frozenAnswerMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "partial",
      "cpuThreads": 4,
      "device": "metal",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 219.881792,
      "loadMs": 655.4910000000001,
      "offloadedLayers": 25,
      "precision": "q4",
      "provider": "llama.cpp Metal",
      "requestedDevice": "metal",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 875.3727920000001,
      "performanceMetric": {
        "name": "charactersPerSecond",
        "value": 40.931083552384365
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "metal",
    "status": "pass",
    "quality": {
      "metric": "independentWhisperWordMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "partial",
      "cpuThreads": 4,
      "device": "metal",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 585.8709170000001,
      "loadMs": 655.4910000000001,
      "offloadedLayers": 25,
      "precision": "q4",
      "provider": "llama.cpp Metal",
      "requestedDevice": "metal",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 1241.3619170000002,
      "performanceMetric": {
        "name": "audioSecondsPerSecond",
        "value": 4.642660902043034
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "metric": "normalizedWordMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "cpu",
      "cpuThreads": 4,
      "device": "cpu",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 1098.537666,
      "loadMs": 6848.559749999999,
      "offloadedLayers": 0,
      "precision": "q4",
      "provider": "llama.cpp CPU",
      "requestedDevice": "cpu",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 7947.097416,
      "performanceMetric": {
        "name": "audioSecondsPerSecond",
        "value": 10.01331164187865
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "metal",
    "status": "pass",
    "quality": {
      "metric": "normalizedWordMatch",
      "score": 1,
      "passed": true
    },
    "runtime": {
      "acceleration": "partial",
      "cpuThreads": 4,
      "device": "metal",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "inferenceMs": 829.374417,
      "loadMs": 762.062084,
      "offloadedLayers": 25,
      "precision": "q4",
      "provider": "llama.cpp Metal",
      "requestedDevice": "metal",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "totalMs": 1591.4365010000001,
      "performanceMetric": {
        "name": "audioSecondsPerSecond",
        "value": 13.26300856950595
      },
      "grade": "good",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "metric": "exactAnswer",
      "score": 1,
      "passed": true,
      "expected": "42"
    },
    "runtime": {
      "acceleration": "cpu",
      "cpuThreads": 4,
      "device": "cpu",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "loadMs": 8682.253875,
      "offloadedLayers": 0,
      "precision": "q4",
      "provider": "llama.cpp CPU",
      "requestedDevice": "cpu",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "inferenceMs": 1118.5454160000008,
      "totalMs": 9800.799291000001,
      "performanceMetric": {
        "name": "exactAnswer",
        "value": 1
      },
      "grade": "excellent",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "metal",
    "status": "pass",
    "quality": {
      "metric": "exactAnswer",
      "score": 1,
      "passed": true,
      "expected": "42"
    },
    "runtime": {
      "acceleration": "partial",
      "cpuThreads": 4,
      "device": "metal",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "loadMs": 549.356709,
      "offloadedLayers": 25,
      "precision": "q4",
      "provider": "llama.cpp Metal",
      "requestedDevice": "metal",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "inferenceMs": 3421.2764590000006,
      "totalMs": 3970.6331680000008,
      "performanceMetric": {
        "name": "exactAnswer",
        "value": 1
      },
      "grade": "excellent",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "cpu",
    "status": "pass",
    "quality": {
      "metric": "exactAnswer",
      "score": 1,
      "passed": true,
      "expected": "Paris"
    },
    "runtime": {
      "acceleration": "cpu",
      "cpuThreads": 4,
      "device": "cpu",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "loadMs": 8096.961625000001,
      "offloadedLayers": 0,
      "precision": "q4",
      "provider": "llama.cpp CPU",
      "requestedDevice": "cpu",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "inferenceMs": 325.3824590000004,
      "totalMs": 8422.344084,
      "performanceMetric": {
        "name": "exactAnswer",
        "value": 1
      },
      "grade": "excellent",
      "platform": "macos",
      "arch": "aarch64"
    }
  },
  {
    "precision": "q4",
    "device": "metal",
    "status": "pass",
    "quality": {
      "metric": "exactAnswer",
      "score": 1,
      "passed": true,
      "expected": "Paris"
    },
    "runtime": {
      "acceleration": "partial",
      "cpuThreads": 4,
      "device": "metal",
      "engine": "llama.cpp",
      "fallback": false,
      "fallbackReason": null,
      "loadMs": 757.386917,
      "offloadedLayers": 25,
      "precision": "q4",
      "provider": "llama.cpp Metal",
      "requestedDevice": "metal",
      "requestedPrecision": "q4",
      "selectionReason": "Exact Q4_0 package with verified layer placement",
      "version": "b12359",
      "artifactRevision": "39670cb5c42e9c4bfe046519ba7ffbe47f9552ad",
      "inferenceMs": 194.50254100000166,
      "totalMs": 951.8894580000017,
      "performanceMetric": {
        "name": "exactAnswer",
        "value": 1
      },
      "grade": "excellent",
      "platform": "macos",
      "arch": "aarch64"
    }
  }
]
Release validation still required
  • Repository-wide production app lint retains seven pre-existing errors in generated vendored DuckDB worker JavaScript; see research/lfm3/production-iteration-10-check-18.log. Scoped catalog lint and current-device production inference, app replay, backup and native source-recovery checks passed.

Something not quite right?

Tell us what happened with Liquid LFM3 Desktop. Loading, results, or the page itself—we’re here for all of it.

Attachments Optional

Up to 3 files · 50 MB each

Browser, OS, CPU/GPU, memory, screen, storage, and network details are included automatically. Model inputs, results, and browsing history are not collected.