GOESB — Global Open Edge Speech Benchmark · pronounced “goose-bee”

Leaderboard

Objective, reproducible benchmarks for local (edge) speech AI. Every result here re-verified its hash and signature before being accepted — see why you can trust the numbers. New here? Read the docs.

This is every verified result, filterable and sortable. Looking for a single best recommendation for your constraints instead? Try the homepage →

Lowest energy Fastest (lowest RTF)Most accurate (lowest WER)

ConcurrencyLoad results only — how RTF/throughput hold up under N simultaneous requests, not accuracy.

Batch Streaming Concurrency

RTF vs Concurrency level

RTF Throughput GPU

Showing the 8 most complete/recent of 28 configurations — 20 more hidden. Narrow with the filters on the left to see them.

amd-epyc-9135 · cpu · vosk/vosk-model-small-en-us-0.15amd-epyc-9135 · cpu · whisper-cpp/whisper-large-v3amd-epyc-9135 · cpu · whisper-cpp/whisper-mediumapple-m1-pro · cpu · faster-whisper/whisper-tinyapple-m1-pro · cpu · vosk/vosk-model-en-us-0.22apple-m1-pro · cpu · vosk/vosk-model-small-en-us-0.15apple-m1-pro · metal · whisper-cpp/whisper-baseapple-m1-pro · metal · whisper-cpp/whisper-tinyClick points to see full details