Unverified50% confidenceBenchmarkExact time
According to data published by Liquid, Audio 2B has an English word error rate of 6.2 and a VoiceBench score of 73; Gemma 4 E4B, with about four times the parameters, scores almost the same.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
11/4/2026
First Seen
Valid until: 2/2/2027
Sources
Running Audio and Vision Locally with a 2B Model: LFM3 Hands-On Revealed
bilibili枫叶边城11/4/2026
Related Entities
Related Claims
UnverifiedThe Gemma speech model is estimated to occupy approximately 5.4G in size based on Q4 quantization, and can run with 8G of VRAM.68% similarUnverifiedOn the tool-calling benchmark, Omni 2B scores 52, Audio 2B scores 58 and Gemma 4 E4B scores 72; tool calling is a clear weakness of LFM3.66% similarUnverifiedOn the LibriSpeech benchmark, Oído's word error rate (WER) is 3.7 (clean) and 8.2 (other), while Whisper tiny.en on a laptop achieves 6.3 and 15.9.65% similarUnverifiedThe author empirically compared Gemma 3 1B and 4B: 1B, in the rename scenario, wrote "the app will resend the shelved event" and told the sender to make the change, and both errors were caught by the checks and fell back to the rule answer; 4B took about 16 seconds on a laptop GPU and accurately stated the customer, amount and specific field.62% similarUnverifiedTerminal-Bench 4.0 has a standard error of ±2.6 points, and Opus 5's 51.8% reproduced as 52.3%, which Anthropic calls noise.61% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1134831API
curl https://kongchang.com/api/v1/knowledge/claims/1134831MCP
get_claim(id=1134831)