Gladia
Real-time audio intelligence API — multilingual transcription, speaker diarization, and voice insights at sub-300ms latency.
Overview
Gladia (Paris, 2022) is a developer-first speech-to-text API — real-time and async transcription across 100+ languages with diarization, PII redaction, and sentiment analysis. OVH Groupe entered exclusive talks to acquire it in June 2026.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.
- $0.75/hr real-time
- 30 real-time / 25 async concurrency
- 100+ languages, diarization
- GDPR/HIPAA/SOC 2 Type 2
- From $0.25/hr real-time
- Custom volume discounts
- Flexible concurrency
- Model-training opt-out
- Unlimited concurrency
- Zero data retention
- Dedicated Slack + AM
- Custom hosting
Use cases
What you can produce with Gladia
- REST and WebSocket API producing JSON transcripts with word-level timestamps and speaker labels
- Real-time audio stream transcription with sub-300ms partial-transcript latency (~700ms for final transcripts)
- Multi-speaker diarisation output identifying and labelling speakers per recording
- Conversation-level enrichments: sentiment scores, named entities, PII-redacted transcripts, and LLM-generated summaries
- Native Python and TypeScript SDKs (v1.0.0, April 2026) plus integrations with Zapier, Make, Pipecat, LiveKit, and Twilio
- Compliance artefacts: SOC 2 Type II, HIPAA, ISO 27001, GDPR data processing agreements
ASEAN Perspective
Gladia in Southeast Asia
Gladia's API is globally accessible from ASEAN with no region-specific restrictions, making it straightforward to integrate into Singapore, Malaysian, or Indonesian tech stacks. However, the company does not yet offer data residency outside EU and US — a meaningful gap for enterprises subject to MAS TRM, Bank Negara, or OJK data localisation rules. Its 100+ language coverage includes Bahasa Indonesia, Malay, Vietnamese, Thai, and Filipino, but accuracy on these languages has not been independently benchmarked at the same rigour as its European and English performance. ASEAN teams building multilingual contact-centre or meeting-intelligence products will find Gladia's code-switching capability particularly valuable for mixed-language calls, though pricing in USD and the absence of regional support channels are minor friction points.
Gladia's biggest 2026 development isn't a model release — it's the pending sale. OVH Groupe entered exclusive acquisition talks on June 11, 2026, to fold Gladia's speech-to-text stack into its sovereign-cloud AI Lab; terms and timeline are still undisclosed. Buyers should factor that uncertainty into any new contract.
On the product itself, Solaria-3 leads the Earnings22 business-audio benchmark (6.4% WER) and still outperforms vanilla Whisper and the hyperscalers on noisy, multilingual call audio. The generous 10-hour free tier and bundled diarization/PII/sentiment features remain a genuine differentiator. Caveats: EU/US-only data residency shuts out latency-sensitive APAC deployments, $20.3M total funding is thin next to AWS and Google, and third-party review volume stays sparse for an API-first product.
What people say
OVH Groupe entered exclusive talks to acquire Gladia on June 11, 2026, aiming to fold its speech-to-text stack into OVH's sovereign-cloud AI Lab — terms undisclosed, no completion date set. For prospective buyers, that's the headline fact right now: evaluate Gladia knowing ownership and roadmap could shift within the next few quarters.
The product itself remains a strong specialist pick. Solaria-3 (launched June 10, 2026) ranks first on the Earnings22 financial-calls benchmark at 6.4% WER and hits 9.6% on real customer call audio, both ahead of AssemblyAI, ElevenLabs, and Deepgram on the same tests. The bundled feature set — diarization via pyannoteAI, PII redaction, sentiment analysis, translation — removes the usual per-add-on nickel-and-diming, and the free tier (10 hours/month, recurring) makes evaluation low-risk. Confirmed customers include HeyGen, Livestorm, Attention, VEED, Aircall, and Recall — not Klarna, despite that name circulating in some secondary write-ups.
Total funding sits at $20.3M, modest next to AWS, Google, and Microsoft's transcription arms. Data residency covers only the EU and US, so ASEAN teams under MAS TRM or PDPA rules will need to accept cross-border transfers for now. Third-party review volume is thin, typical for an API-first product with no consumer app, which makes the OVH tie-up the single most important data point for anyone evaluating Gladia this quarter.
Summary of public user & expert reviews, compiled by RECATOOLS.
Notable facts
- Gladia founder Jean-Louis Quéguiner built the company after frustration that transcription APIs couldn't handle his French accent — a deeply personal origin story for a multilingual AI product.
- Gladia's Whisper-Zero model reduces hallucinations by up to 99.9% compared to vanilla OpenAI Whisper — hallucinations in transcription APIs cause real business errors like mis-quoted prices or names in sales calls.
- The platform has transcribed over 2 billion minutes of audio as of mid-2026, roughly equivalent to 3,800 years of continuous speech.
- Gladia was part of Sequoia Capital's first-ever Sequoia Arc programme in France, giving it early backing from one of Silicon Valley's most storied VC firms.
Frequently asked questions
About this listing
This entry was compiled from publicly available data including Gladia's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with Gladia unless explicitly stated.
Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.
For the latest details, please refer to Gladia directly →
Spotted something out of date? Suggest an update →
More in Video & Audio
