Gladia

The buyer Gladia targets is technical. Gladia's own sales form lists CCaaS, CPaaS, BPO, telephony, vocal agent, note taker and content or media as the use cases teams arrive with, which matches an audience of engineers embedding transcription into a product rather than staff transcribing their own calls. Gladia advertises support for 100+ languages with native code-switching so a sentence that shifts language mid-flow does not break structure or timestamps, and Gladia states that latency, pricing and SLA stay consistent across every supported language.

Day-to-day work with Gladia usually begins in the playground or developer console, then moves to the official Python, Node.js and WebSocket reference clients. Gladia returns word-level timestamps, speaker diarization with speaker-level confidence, named entity recognition, custom vocabulary spelling and context-aware punctuation, casing and numeral formatting. Gladia also ships an Audio-to-LLM contract that produces summaries, action items, entity extraction and sentiment without a second-hop LLM call, and Gladia delivers async results by webhook as soon as processing completes.

Model selection is a real decision inside Gladia. Solaria-1 is the broad Gladia model for 100+ languages, code-switching, formal institutional speech and real-time streaming with sub-103ms partials, while Solaria-3 is the newer Gladia production model tuned for noisy business audio in English, French, German, Spanish and Italian. Gladia publishes Solaria-3 word error rates openly, including 9.6% on real customer English audio and 6.4% on Earnings22 cleaned, and Gladia also publishes regressions of 2.9% on VoxPopuli cleaned and 8.0% on Multilingual LibriSpeech against Solaria-1.

Packaging on Gladia is usage-based rather than seat-based. The Gladia Starter plan is pay-as-you-go at $0.61 per hour async and $0.75 per hour real-time with 30 real-time and 25 async concurrent requests, and Gladia grants 50€ in free credits that the Gladia FAQ estimates at roughly 80+ hours of pre-recorded or 60+ hours of real-time transcription. Growth on Gladia drops rates to as low as $0.20 per hour async and $0.25 per hour real-time, described by Gladia as up to 67% less than Starter, in exchange for an upfront commitment. Enterprise on Gladia is an annual contract adding custom models, fine-tuning, debundled pricing, zero data retention, dedicated Slack support and custom hosting, and Gladia bills everything from a prepaid wallet with manual or automatic Stripe top-ups.

Limitations are worth reading closely before committing to Gladia. The 50€ grant from Gladia is one-time with no monthly reset, Gladia publishes only floor rates for Growth rather than the full commit ladder, and Gladia lists the 99.9% uptime SLA, priority processing queue and dedicated infrastructure as higher-tier items rather than Starter defaults. Solaria-3 inside Gladia is async only at launch and optimised for five European languages, so long-tail language work and streaming still route to Solaria-1. There is also a documented tension: the Gladia homepage promises "No training on your audio — contractual, not marketing, verifiable in the DPA", while the Gladia Inc. general terms of use grant Gladia a perpetual licence to use uploaded Data to improve the Services "including without limitation model training", and the Gladia pricing table presents model training opt-out as an automatic Growth feature and a default Enterprise feature.

Against category peers, Gladia competes with Deepgram, AssemblyAI, Speechmatics, ElevenLabs Scribe and OpenAI Whisper deployments, and Gladia leans on EU headquarters, 100% EU data residency, GDPR, HIPAA, SOC 2 Type II and ISO 27001 documentation as the differentiator for European procurement. Gladia also publishes multi-provider WER tables that include unfavourable rows, which is a more auditable posture than a single headline accuracy claim. Teams that want a seat-priced meeting recorder with a polished end-user interface will find Gladia the wrong shape, because Gladia stops at the API, the console and the integrations that feed it.