Cekura Bench launches voice AI benchmark spanning 50+ STT, TTS and speech-to-speech models
Scobleizer · x · 2026-11-06
Cekura Bench released a comprehensive voice AI benchmark suite covering speech-to-speech, STT, TTS models and voice agent platforms, testing 50+ models from OpenAI, Google, Nvidia, xAI, Inworld, Phonic, Fish Audio and more, using recorded real-human task conversations.
Key results:
- Speech-to-speech: Grok Voice 3.0 most reliable (42.9%), Grok Voice Think Fast 2.0 fastest reply (1.44s), GPT Realtime 2.1 Mini cheapest ($0.020); Phonic v1 tops agent reliability (90.2%)
- Turn Bench: Gemini 3.1 Flash Live best turn-taking (88.4%), Higgs Realtime best in heavy noise (87.2%)
- STT: Fish Audio Transcribe Stream 1 most accurate (2.21% WER), Soniox STT-RT v5 fastest (538ms)
- TTS: Inworld TTS-2 Flash lowest non-streaming latency (46ms), Soniox TTS RT v2 cheapest
- Voice platforms: PolyAI most reliable (87.8%), ElevenLabs fastest reply (1.27s), Bland best voice quality
The team is open to collaborating with more labs on the benchmarks.
Related event: Cekura Launches First Public Voice AI Benchmark Leaderboard(2 posts)→
More from Models
- GPT-6 Cyber 'lightyears ahead of everything', says researcher after OpenAI early-adopter reveal — Eric_Wallace_ · 2026-11-06
- Fastino open-sources GLiDE: five decision models (0.4B-27B) claiming a new Pareto frontier — HowDevelop · 2026-11-06
- V0.5 model released with technical report, Hugging Face weights and inference code — rohanpaul_ai · 2026-11-06
- Clairvoyance launches Local AI benchmark site: 31 models × 15 hardware configs — draginol · 2026-11-06
- webAI Claims 12B Model Beats Gemini and Claude, But the Benchmark Claims Raise Eyebrows — eurie_kim · 2026-11-06
- User cancels OpenAI over Codex quota change, says Claude did a week's work in 4.5 hours — LethalLuck7 · 2026-11-06