Unverified50% confidenceBenchmarkExact time
Whisper has been iterated to the large-v3 version (November 2023), with an average word error rate (WER) on the Fleurs benchmark reduced to approximately below 10%.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
9/9/2026
First Seen
Valid until: 12/8/2026
Sources
Related Entities
Related Claims
UnverifiedOn the LibriSpeech test-clean benchmark, Whisper large-v3's WER has dropped to around 2%, approaching or in some cases surpassing human annotators (approximately 4–5%).78% similarVerifiedWhisper is available in multiple sizes ranging from tiny (39M parameters) to large-v3 (1.5B parameters).63% similarUnverifiedOn the SWE-Bench Multilingual benchmark, HY3 scored 75.8, slightly behind DeepSeek V4 Pro's 76.2.58% similarUnverifiedEarly testing on the Opus model showed an exact string recognition rate of 0 out of 15, whereas after the release of Fable 5 the same test yielded 13 out of 15.56% similarUnverifiedWhisper large-v3 是目前精度最高的版本，约15亿参数，采用 MIT License53% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/884488API
curl https://kongchang.com/api/v1/knowledge/claims/884488MCP
get_claim(id=884488)