Proven performance, backed by data
Shunyalabs’ transcription leads the way — producing 48% fewer errors than the next best model.
Highest Word Accuracy Rate
100
95
90
85
98.57%
98.57%
97.29%
97.29%
96.60%
96.60%
97.06%
97.06%
96.62%
96.62%
96.88%
96.88%
Pingala-v1-universal
Shunya Labs
canary-qwen-2.5b
NVIDIA
granite-speech-3.3-8b
IBM
Phi-4-multimodal-instruct
Microsoft
parakeet-tdt-0.6b
NVIDIA
canary-1b-flash
NVIDIA
Lowest Total Word Error Rate
3.10
%
%
5.63
%
%
5.74
%
%
6.02
%
%
6.05
%
%
6.35
%
%
8
6
4
2
0
Pingala-v1-universal
Shunya Labs
canary-qwen-2.5b
NVIDIA
granite-speech-3.3-8b
IBM
Phi-4-multimodal-instruct
Microsoft
parakeet-tdt-0.6b
NVIDIA
canary-1b-flash
NVIDIA
Milestones
0ms
round-trip latency in production
0
languages & dialects
0%
GPU cost saved for a Fortune 100 logistics client
0.00%
Word error rate