# LFM3-Audio-2B Benchmark Scores & Performance

> LFM3-Audio-2B has Liquid LFM3 audio results results. Voice evidence is display-only and does not produce a text-model score.

Canonical page: https://benchlm.ai/models/lfm3-audio-2b

Last updated: October 27, 2026

## Model Details

| Property | Value |
|----------|-------|
| Creator | LiquidAI |
| Source Type | Open Weight |
| Reasoning Type | Reasoning |
| Context Window | 32K |
| Official model card | [LiquidAI model card](https://huggingface.co/LiquidAI/LFM3-Audio-2B/blob/9d49701c372d6011d900f6abb40c6b296fb706dd/README.md) |
| Overall Score | Not scored (voice evidence only) |
| Overall Rank | Unranked |

## Family & Coverage

- Family: LFM3
- Variant: base
- Benchmarks covered: 0 of 716
- Sibling models: [LFM3-1.7B](/models/lfm3-1-7b), [LFM3-1.7B-Base](/models/lfm3-1-7b-base), [LFM3-Omni-2B](/models/lfm3-omni-2b)
- Coverage note: Speech and voice results appear below; no weighted text-model benchmark rows are stored.

## Speech and voice evidence

[Liquid AI LFM3 audio and omni evaluation](/voice-benchmarks/liquid-lfm3-audio-omni-evaluation)

Liquid AI measured every result, including the competitor columns not carried here, with each model’s own chat template and greedy text decoding. English ASR averages the ESB / Open ASR suite; multilingual ASR averages FLEURS over nine languages, with CER for Chinese, Japanese and Korean. VoiceBench is the mean of 9 subsets with the three 1-5 judge scores multiplied by 20. The tool-calling score is an internal evaluation without a public task set. These results do not enter text-model or voice-agent rankings.

| Metric | Result |
| --- | --- |
| English ASR WER (ESB / Open ASR) | 6.23% |
| Multilingual ASR WER / CER (FLEURS, 9 languages) | 6.03% |
| English TTS UTMOSv2 | 3.83 |
| VoiceBench | 73.13 |
| Audio tool calling (internal) | 58.50 |
| MMAR audio understanding | 50.60 |

Liquid AI launch post and model cards retrieved October 27, 2026. Provider-reported; error rates are better lower. Voice results are display-only.

- [Evaluation source](https://www.liquid.ai/blog/lfm3)

LFM Open License v1.0 weights on Hugging Face with GGUF, ONNX and MLX exports, released October 27, 2026 with a 32K context. Audio comes in at 16 kHz and speech goes out at 24 kHz across 19 speech languages and locales. Liquid reports time to first token as low as 50 ms and LiquidCodec decoding at about 87x real time on one Apple M4 Max CPU thread; text-only answers can reason first.

## Other LiquidAI Models

- [LFM2.5-24B-A2B](/models/lfm2-5-24b-a2b) - Score: 36.99
- [LFM2.5-2.6B](/models/lfm2-5-2-6b) - Score: 31.4
- [LFM2.5-8B-A1B](/models/lfm2-5-8b-a1b) - Score: 30.31
- [LFM3-1.7B](/models/lfm3-1-7b) - Score: 22.99
- [LFM2-24B-A2B](/models/lfm2-24b-a2b) - Score: 20.74
- [LFM2.5-1.2B-Thinking](/models/lfm2-5-1-2b-thinking) - Score: 19.5
- [LFM2.5-1.2B-Instruct](/models/lfm2-5-1-2b-instruct) - Score: 19.25
- [LFM2.5-350M](/models/lfm2-5-350m) - Score: not computed
- [LFM2.5-VL-450M](/models/lfm2-5-vl-450m) - Score: not computed
- [LFM3-VL-2B](/models/lfm3-vl-2b) - Score: not computed
- [LFM2.5-230M](/models/lfm2-5-230m) - Score: not computed
- [LFM2.5-VL-3B](/models/lfm2-5-vl-3b) - Score: not computed
- [LFM2.5-VL-1.6B-Extract](/models/lfm2-5-vl-1-6b-extract) - Score: not computed
- [LFM2.5-VL-450M-Extract](/models/lfm2-5-vl-450m-extract) - Score: not computed
- [d1-2B](/models/d1-2b) - Score: not computed
- [d1-Omni-2B](/models/d1-omni-2b) - Score: not computed
- [LFM3-1.7B-Base](/models/lfm3-1-7b-base) - Score: not computed
- [LFM3-Omni-2B](/models/lfm3-omni-2b) - Score: not computed
- [d1-3B](/models/d1-3b) - Score: not computed
- [d1-omni-600M](/models/d1-omni-600m) - Score: not computed
- [LFM2.5-ColBERT-350M](/models/lfm2-5-colbert-350m) - Score: not computed
- [LFM2.5-Embedding-350M](/models/lfm2-5-embedding-350m) - Score: not computed
- [LFM2.5-1.2B-JP-202606](/models/lfm2-5-1-2b-jp-202606) - Score: not computed