Gradium AI Releases New Default TTS Model: 81.0% Hard-Case Pass Rate at 216 ms Time-to-First-Audio

Gradium AI released a new text-to-speech model that converts written text into spoken audio, now set as the default option across its service as of August 31, 2026. The model matters because voice agents often fail on critical information like phone numbers, email addresses, and reference codes that callers need to hear accurately, and this new model achieved an 81.0% human-rated success rate on difficult test cases across five languages, outperforming competing models from Cartesia, ElevenLabs, Fish Audio, and Inworld. The model also generates audio quickly, producing the first audio output in 216 milliseconds with minimal variation between attempts, ranking it competitively among tested alternatives. Existing users require no action to receive these improvements, and Gradium has published its 500-sentence evaluation dataset openly on Hugging Face so others can verify the results.