Prerecorded speech recognition for English, French, German, Spanish and Italian business audio. It does not support live streaming or mid-file language switching.
Exactly five supported languages. The model guide allows several candidates for detection, while the API introduction says pass exactly one; a single explicit language is the clearest common contract.
Does not retrain the model. The guide recommends focused lists and warns about false-positive replacements; no hard vocabulary-list limit is published.
Supported base versions, dataset requirements, training limits, custom artifact selectors and pricing are not specified. No self-service fine-tuning API is inferred.
Access & pricing
Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.
Top-level model parameter; omission defaults to solaria-1.
Access
POST https://api.gladia.io/v2/pre-recorded with audio_url, model and options; upload first or provide an external URL. Receive callback/webhook or poll results when retention settings allow.
Regions
EU and US regional infrastructure; verify account routing for this transport
Prices retain their published currencies and billing units below.
Published billing units · original currency · observed 13 Oct 2026
Category
Usage
Rate
Unit
Speech recognition
Starter async transcription1
$0.61
audio hour
Speech recognition
Growth · as low as async transcription2
$0.20
audio hour
Rate conditions
Published pay-as-you-go rate through prepaid credit wallet. Same plan rates for supported Solaria models. Core audio-intelligence features are bundled; Audio-to-LLM uses external model rates plus an unspecified platform fee. Different-content stereo channels are billed as two audios; identical-content channels are billed once. Custom Enterprise bundling and prices require a quote.
Advertised lower bound requiring upfront commitment and volume agreement; not a universal pay-as-you-go price. Same plan rates for supported Solaria models. Core audio-intelligence features are bundled; Audio-to-LLM uses external model rates plus an unspecified platform fee. Different-content stereo channels are billed as two audios; identical-content channels are billed once. Custom Enterprise bundling and prices require a quote.
Paid defaults: audio and transcripts 3 weeks, API metadata 1 year, logs an unspecified limited period. Free defaults retain all four data types for 1 year. Enterprise custom retention value 0 means deletion within 24 hours; separate ZDR means ephemeral processing and no later retrieval. ZDR disables file upload and polling: supply an external audio URL and receive callbacks. The compliance hub says no training without explicit agreement; a pricing blog says Starter training may be enabled by default, so confirm the applicable account agreement.
Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.
01 / INITIAL INPUTProcess & retain
The system processes eligible input and retains reusable state.
02 / LATER REQUESTCheck available state
Reuse depends on matching input, session state and the cache policy.
03 / CACHE HITReuse & generate
Reusing state can reduce repeated work. Any charges follow the billing rules below.
Gladia API · async
solaria-3
Not verified
No model-specific discounted prompt-cache contract verified. Streaming buffers, stored transcripts and saved custom vocabulary are not token caching.
How it works
Not collected
No model-specific discounted prompt-cache contract verified. Streaming buffers, stored transcripts and saved custom vocabulary are not token caching.
Minimum reusable input
Not collected
No model-specific discounted prompt-cache contract verified. Streaming buffers, stored transcripts and saved custom vocabulary are not token caching.
Lifetime & refresh
Not collected
No model-specific discounted prompt-cache contract verified. Streaming buffers, stored transcripts and saved custom vocabulary are not token caching.
How to enable it
Not collected
No model-specific discounted prompt-cache contract verified. Streaming buffers, stored transcripts and saved custom vocabulary are not token caching.
Usage fields
Not collected
No model-specific discounted prompt-cache contract verified. Streaming buffers, stored transcripts and saved custom vocabulary are not token caching.
Billing rules
Not collected
No model-specific discounted prompt-cache contract verified. Streaming buffers, stored transcripts and saved custom vocabulary are not token caching.
Limits & data boundaries
Not collected
No model-specific discounted prompt-cache contract verified. Streaming buffers, stored transcripts and saved custom vocabulary are not token caching.
No verified cache rate for this endpoint. An empty rate is not a zero charge.
What the measurements say
Gladia internal English production audio
Gladia · provider-reported
9.6% WER
Provider evaluation of human-annotated customer meetings and support calls, published June 10, 2026. Private dataset; not independently reproducible and not all-language accuracy.
Hugging Face Open ASR Leaderboard · independent benchmark
5.94% WER
Benchmark-owner pinned commit d3c0e9a07b475a64afb56254e8c35b584905af32; exact Earnings22-Cleaned-AA-chunked WER column. Lower is better; this is not the original Earnings22 score.
Provider-run production APIs with default settings and Whisper text normalization; lower WER is better. Switchboard — Conversational Speech. No independent replication or overall model ranking is claimed.
Provider-run production APIs with default settings and Whisper text normalization; lower WER is better. Common Voice 24 — Clean Audio (Multilingual). No independent replication or overall model ranking is claimed.
Provider-run production APIs with default settings and Whisper text normalization; lower WER is better. VoxPopuli — Formal Discourse (European Parliament). No independent replication or overall model ranking is claimed.
Provider-run production APIs with default settings and Whisper text normalization; lower WER is better. Multilingual LibriSpeech — Audiobooks (5 languages, average). No independent replication or overall model ranking is claimed.
Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.
The record behind the model
Why it is included
Gladia describes Solaria-1 and Solaria-3 as its internally trained production speech models. Its licensed service reserves model/algorithm rights and exposes inference rather than model weights. Solaria-1’s documented Whisper-derived ensemble ancestry does not turn the proprietary combined service into a public-weight Whisper release.
