Gladia

Gladia is a Paris-based audio-AI company (founded 2022) offering a speech-to-text API positioned as production infrastructure for voice products: meeting assistants, contact centers, voice agents, and call analytics. Its core differentiator is multilingual breadth and real-world robustness. The current flagship models, Solaria-1 and the newer Solaria-3, support 100+ languages with native code-switching, automatic language detection, speaker diarization, translation, sentiment analysis, and PII redaction bundled into the base price rather than charged as separate add-ons. The company raised a $16M (€14.5M) Series A in October 2024 led by XAnge, explicitly to push into real-time/streaming transcription, which it frames as the next frontier beyond Whisper-style batch transcription.

On accuracy, Gladia publishes third-party-dataset benchmarks where Solaria-3 leads: 6.4% WER on Earnings22 business audio (ahead of AssemblyAI 6.9%, ElevenLabs 7.7%, Speechmatics 7.8%, Deepgram 12.0%) and the only model under 35% WER on the noisy Switchboard telephone set. For real-time, it advertises ~270ms average partial latency and sub-300ms finals. These are vendor-run benchmarks, so they should be read as favorable-but-credible rather than independent. Pricing is aggressive and transparent: a genuinely usable 10-hour/month free tier, pay-as-you-go Starter at $0.61/hr async and $0.75/hr real-time, and committed Growth pricing dropping to roughly $0.20/hr async, undercutting Deepgram, AssemblyAI, and Speechmatics on list price while including audio-intelligence features.

Gladia's sweet spot is European/multilingual, telephony-heavy, and EU-compliance-sensitive workloads (GDPR/HIPAA/SOC 2 Type 2, zero-data-retention and EU hosting on Enterprise). The main tradeoffs are the lighter ecosystem versus US incumbents (fewer official SDKs, a smaller community, less third-party review volume), some real-time tuning rough edges (e.g. a 5-second minimum endpointing window that frustrates dictation use cases), and the fact that most published accuracy claims originate from Gladia itself. For teams that need many languages, clean diarization, and competitive per-hour cost without paying à la carte for features, it's a strong contender; teams wanting the deepest tooling/community or fully independent benchmarks may still default to Deepgram or AssemblyAI.