Gladia

What Is Gladia?

Teams building voice products in Europe have historically reached for Deepgram or AssemblyAI by default, mostly because those were the names that showed up first in a Google search. Gladia gives them a European-built alternative that does not ask for a trade-off on accuracy to get EU data residency. Founded in 2022 and based in Paris, Gladia builds a single API that turns raw audio — phone calls, meeting recordings, voice-agent streams — into transcripts, speaker labels, sentiment scores, and summaries.

The company's positioning is deliberately narrow: audio infrastructure, not a general-purpose AI platform. That focus shows in the product. Gladia supports real-time transcription with sub-300ms latency and asynchronous batch processing for longer recordings, both running on its own Solaria model family. Solaria-3, launched in June 2026, was tuned specifically for business audio — contact-centre calls, sales recordings, and meetings with overlapping speech and background noise — rather than the cleaner studio audio that many benchmarks favour.

Gladia serves developers building contact-centre analytics, meeting assistants, subtitling tools, and voice agents. Customers referenced by the company include HeyGen, Livestorm, and Recall.ai. Its differentiator against Deepgram and AssemblyAI is twofold: broader language coverage with native code-switching, and a French legal entity that makes EU data residency a straightforward configuration rather than a workaround.

Key Features

Solaria Speech Recognition

Solaria-3 is Gladia's core transcription model, and the company publishes open benchmarking data showing it ranking competitively on business-audio accuracy against Deepgram and AssemblyAI. The model handles code-switching mid-sentence across its 100+ supported languages, which matters for multilingual European call centres where an agent and customer might mix languages within a single conversation. Gladia claims 42 languages that competitors do not cover at all.

Built-In Audio Intelligence

Where many STT APIs stop at a transcript, Gladia layers analysis on top: speaker diarization, sentiment analysis, named entity recognition, topic detection, summarisation, and translation. For a contact-centre analytics product, this collapses what would otherwise be a multi-vendor pipeline — transcription from one provider, NLP from another — into a single API call. Custom vocabulary support also helps with industry-specific terminology that generic models mishear.

Real-Time and Async Modes

Real-time transcription targets sub-300ms latency for live captioning and voice-agent use cases, while async mode adds hallucination filtering for cleaner output on long-form recordings. Both modes ship with the same diarization and language-detection features baked into the base price, rather than gating them behind an add-on fee — a contrast with providers that charge extra for diarization.

Integration Paths

Gladia offers direct REST API access with Python and Node.js SDKs, plus native plugins for LiveKit, Twilio, Pipecat, and Retell, which cover most voice-agent and telephony stacks out of the box. For teams that want to prototype without writing code, Gladia also connects through Zapier, Make, and n8n — the latter useful for privacy-sensitive workloads that need to stay self-hosted.

Pricing

Gladia's pricing is transparent and per-hour rather than per-minute-with-hidden-fees. The Starter tier runs $0.61 per hour for async transcription and $0.75 per hour for real-time, with diarization, language detection, and the full 100+ language set included at no extra charge. New accounts get a one-time €50 credit grant, equivalent to roughly 80 hours of async processing, instead of a recurring free monthly allowance.

The Growth tier drops per-hour rates to roughly $0.20-0.25 with an upfront volume commitment. That's a meaningful discount for teams with predictable call volumes, though it's a harder sell for anyone whose traffic swings month to month. Enterprise pricing is custom and adds unlimited concurrency, zero data retention, and custom hosting. It also throws in a dedicated Slack channel with an account manager, features that show up as line items on procurement checklists for regulated industries.

Compared to Deepgram's and AssemblyAI's published rate cards, Gladia sits in a similar per-hour range on the Starter tier, with the real difference showing up in the audio-intelligence bundle rather than the raw transcription price.

EU Compliance & Privacy

Gladia SAS is registered under RCS Rennes and operates from Paris, putting it squarely under EU jurisdiction. The company offers EU data residency as a configurable option and holds GDPR, SOC 2 Type II, HIPAA, and ISO 27001 coverage. That compliance stack is aimed squarely at the contact-centre and healthcare-adjacent customers who make up a meaningful share of its base. Enterprise customers can also enable zero-retention processing, meaning audio is discarded immediately after transcription rather than stored.

One development worth flagging plainly: OVH Groupe, the French cloud infrastructure group, announced in June 2026 that it had entered exclusive negotiations to acquire Gladia as part of a push into sovereign voice AI. The deal had not closed as of this review, and financial terms were not disclosed. Prospective customers should treat Gladia's ownership structure as potentially in flux over the near term. It's worth asking directly about data governance commitments during procurement, since a change of control can affect existing contracts.

Who It's Best For

Contact-centre and CCaaS builders who need transcription plus analytics — sentiment, topic detection, summarisation — without stitching together separate vendors get the clearest value from Gladia's bundled approach.

Voice-agent developers using LiveKit, Twilio, or Pipecat benefit from native plugin support that skips custom integration work most competitors require.

EU-regulated teams in finance or healthcare that need contractual EU data residency and HIPAA coverage should shortlist Gladia alongside Deepgram, since both offer comparable compliance tooling but only one is a French legal entity.

Multilingual European operations with code-switching customers — a French-Arabic call centre, for instance — get language coverage that neither Deepgram nor AssemblyAI matches as broadly.

Teams that need years of proven uptime at massive scale, or a large existing library of community-built integrations, may still lean toward the more established Deepgram, at least until the OVH acquisition situation resolves.

The Verdict

Gladia earns its place on a genuine Deepgram-and-AssemblyAI shortlist rather than as a token EU alternative. Solaria-3's accuracy on messy business audio, the bundled audio-intelligence stack, and transparent per-hour pricing make a strong technical case on their own. Timing is what holds it back from an unreserved recommendation right now. The pending OVH acquisition adds a layer of uncertainty to long-term roadmap and ownership questions that a fast-growing Series A startup would not normally carry. Buyers comfortable revisiting that question in six to twelve months will find a capable, well-priced European alternative today.

Frequently Asked Questions

Is Gladia GDPR compliant?

Yes. Gladia is a French company (Gladia SAS, registered in Cesson-Sévigné, operating from Paris) that offers an EU data residency option and holds GDPR, SOC 2 Type II, HIPAA, and ISO 27001 coverage. Data is not used for model training by default on Growth and Enterprise plans; Starter customers must request the opt-out.

Is Gladia being acquired by OVH?

OVH Groupe announced exclusive negotiations to acquire Gladia in June 2026 to build sovereign voice AI infrastructure. As of publication the deal has not closed and financial terms were not disclosed, so Gladia continues to operate independently pending completion.

How does Gladia's pricing compare to Deepgram and AssemblyAI?

Gladia's Starter tier starts at $0.61 per hour for async transcription and $0.75 per hour for real-time, with diarization and language detection included at no extra cost. Growth-tier volume commitments bring per-hour rates down to roughly $0.20-0.25, competitive with both US rivals.

What languages does Gladia support?

Gladia covers more than 100 languages with native mid-sentence code-switching, including a number of languages its competitors do not offer. The Solaria-3 model was specifically tuned to improve accuracy in English, French, German, Spanish, and Italian business audio.

Does Gladia offer a free trial?

Gladia grants €50 in free credit on sign-up, equivalent to roughly 80 hours of async or 60 hours of real-time transcription, rather than a time-limited trial. There is no separate trial-day countdown; the credit simply runs out once used.