Gladia
NewGladia is a speech-to-text API for recorded and live audio, with Solaria models covering 100+ languages and code-switching, plus diarization, subtitles, translation, summaries and entity extraction as per-hour add-ons.
Gladia
Explore features, practical uses and pricing below.
Gladia is a speech-to-text API for developers who need transcripts from recorded files or live audio streams. It charges for transcription by the hour and offers optional audio intelligence, such as translation, summaries, sentiment and named entities, returned with the transcript in the same request. Much of its pitch is about multilingual audio and speakers who switch languages mid-sentence.
Solaria-1 and Solaria-3 do different jobs
Gladia runs two models, and choosing between them comes first. Solaria-1 is the generalist: it covers 100+ languages, handles code-switching, and is the only model for real-time streaming. Solaria-3 is newer, async only, takes exactly one language per job, and is optimized for English, French, German, Spanish and Italian. The Solaria-3 page reports, on Gladia's own benchmarks, a 9.6% word error rate on an internal set of customer calls, 26% better than Solaria-1. The same page shows Solaria-3 doing worse on clean read speech and formal institutional recordings, and Gladia advises staying on Solaria-1 for those.
What a pre-recorded job looks like
The pre-recorded quickstart has three steps: upload the file to /v2/upload, create a job at /v2/pre-recorded with the returned audio URL and your options, then poll the job or wait for a webhook or callback. Options are set per request:
- Diarization, with optional speaker counts, to label who said what.
- Custom vocabulary for product names, jargon and people.
- Translation into target languages, with context and an informal-tone setting.
- Subtitles exported as SRT or VTT.
- PII redaction and sentiment analysis, which can pick out up to 25 emotions.
Live transcription for calls and voice agents
For streaming, you first POST a configuration to /v2/live, including encoding, sample rate, bit depth and channels, and get back a WebSocket URL. Audio goes in as binary frames, partial and final transcripts come back sharing an utterance ID, and a stop message triggers post-processing before the socket closes. A single live session can last up to 3 hours. Gladia lists ready-made integrations with Pipecat, LiveKit, Twilio and Retell.
Example: a multilingual support queue
A support team in Brussels records calls in French, Dutch and English, often within the same call. They send recordings through Solaria-1 with diarization set to two speakers and a custom vocabulary of product names. Translation into English is switched on so managers can read every call, and named entity recognition pulls account names and dates into the ticket. Calls from the English-only line could go to Solaria-3 once the team has checked it on a sample of their own recordings.
Who Gladia suits and what to verify
Gladia suits product teams building meeting tools, call analytics, captioning and voice agents, particularly with European languages. It's an API, so someone who simply wants to upload a file and read a transcript will be better served by an app such as TurboScribe. Check two things before committing. First, data residency: the homepage promises EU residency, while its own comparison table limits that to Enterprise. Second, training: on self-serve plans, model training on your data is on by default, though you can opt out at no cost from your profile. Other developer options are in our speech to text API category.
Hourly rates and add-ons
The pricing page lists a self-serve plan starting at €0.16 an hour for async audio on Solaria-3 and €0.27 on Solaria-1 at the lowest volume tier, with rates falling as monthly volume rises, and €0.68 an hour for real time. New accounts get €50 in credits. Add-ons are priced per hour, from €0.02 for sentiment to €0.07 for entity recognition or PII redaction. Self-serve accounts can run 25 async jobs and 30 live sessions at once. Enterprise removes those limits and adds zero data retention, custom hosting and fine-tuning.
Gladia questions from developers
Can Gladia transcribe in real time?
Yes, with Solaria-1 over WebSocket. Solaria-3 is async only for now.
What if speakers switch languages?
Use Solaria-1, which supports code-switching. Solaria-3 jobs take exactly one language in the request.
Is my audio used to train models?
On self-serve plans, training is on by default with a free opt-out. On Enterprise it's off by default.