Rumble-2-Flash is now open weights, Apache 2.0.
Engineering insights from the Reson8 team
Our most accurate speech-to-text model, now live in 50+ languages. Lowest streaming WER and p95 latency on English, and ahead of medical-specific models.
A voice agent for more than 1,000 restaurants that no longer cuts callers off mid-sentence and gets contact details right the first time, on the call.
Our smallest, fastest and most affordable model yet. It understands 40+ languages, starting at €0.01 per hour of audio, and it's live today.
What to measure, how to make your test set sound like production, why reference transcripts are less reliable than you think, and how much audio you need.
Medication names, dosages and readings that come back right in four languages, with 62% lower word error rate on mixed-language audio and transcripts five times faster.
#82 on Sifted's ranking of Europe's most promising AI startups in 2026. Founded this year, and at 10 people the leanest company on the list.
End-of-turn detection that lets candidates finish their thought, with 50% longer utterances and full rollout 15 days after the first meeting.
No cloud provider in the processing chain, no audio retained, and inference on hardware Reson8 owns and runs in a Dutch data center.
Word-level timestamps precise enough to display on and to edit on, tuned to the way VEED benchmarks every provider it runs.
An EU region only answers where your audio sits. Owning the hardware in the Netherlands answers who owns it, whose law applies and what is kept.
Accurate transcription of adviser meetings, with Reson8 self-hosted in Obsidian's own AWS environment.
Bias the model toward the structured identifiers your transcripts depend on — order codes, invoice numbers, licence plates — and cut identifier error by up to 81%.
Eager, accurate turn detection for voice agents, with native barge-in — so your agent knows when to listen, when to speak, and when to get out of the way.
Two frontier speech models trained with reinforcement learning, served via Violin — our inference engine built for speech.