Gladia

Gladia is a solid pick for developers who prioritize language breadth and bundled AI features. With 100+ languages in real-time and automatic code-switching, it handles multilingual conversations better than most rivals. The built-in speaker diarization, sentiment, entity extraction, and summarization save you from stitching multiple providers together — and they don't cost extra.That said, the shift to credit-based billing (announced July 2026) is a change. If you're used to paying per hour of audio with no upfront commitment, the new wallet and auto-top-up model requires more planning. At $0.61/hr async and $0.75/hr real-time on the Starter plan, pricing is comparable to Deepgram, but the Growth plan drops to as low as $0.20/hr async with a volume commitment.Enterprise users get custom models and fine-tuning, but there's no on-premise option — if you need full offline deployment, look elsewhere.Compared to AssemblyAI, Gladia offers more languages in real-time and doesn't charge for speaker diarization. Against Deepgram, the multilingual support is wider, but Deepgram's nova-2 model may edge ahead on pure English WER. Test on your own audio before committing.In practice, we'd reach for Gladia when building a voice agent or meeting assistant that must handle multiple languages out of the box. The free playground (10 hours) lets you evaluate without a credit card. Where it bites: the credit model adds friction for low-volume hobbyists, and you can't fine-tune models on the Starter tier.