Models with data for all selected languages

FLEURS · English · average error, lowest first. Languages written without spaces use character error rate (CER). * marks a vendor report.

Models with FLEURS data for every selected language, lowest average error first.
#	Model	Average WER	English
1	GPT TranscribeOpenAI	2.40%	2.40%^*
2	GPT-4o TranscribeOpenAI	2.40%	2.40%^*
3	MAI Transcribe 1Microsoft	2.70%	2.70%^*
4	Scribe v2ElevenLabs	2.70%	2.70%^*
5	Qwen3 ASR Flash 1208Alibaba	2.72%	2.72%^*
6	Gemini 2.5 ProGoogle	2.94%	2.94%^*
7	Voxtral Mini Transcribe 2Mistral	3.30%	3.30%^*
8	Qwen3 ASR 1.7BAlibaba	3.35%	3.35%^*
9	Gemini 3.1 FlashGoogle	3.60%	3.60%^*
10	GPT-4o Mini TranscribeOpenAI	3.70%	3.70%^*
11	Whisper Large v3OpenAI	4.20%	4.20%^*
12	Qwen3 ASR 0.6BAlibaba	4.39%	4.39%^*
13	Canary 1B V2NVIDIA	4.50%	4.50%^*
14	Gemini 2.5 FlashGoogle	4.60%	4.60%^*
15	Parakeet TDT 0.6B V3NVIDIA	4.85%	4.85%^*
16	Voxtral Mini Transcribe RealtimeStreamingMistral	5.10%	5.10%^*
17	Fun ASR MLT NanoAlibaba	5.49%	5.49%^*
18	Doubao ASRByteDance	6.31%	6.31%^*

Every language on FLEURS

Selected languages first. Languages with fewer than 2 models are hidden unless you search for them.

Error rate, %

*: vendor report±: sources differ by more than 10%