--- format: "aidr-story-markdown/v1" id: "dcc32713a8001b149829f0dae57ecfdb33ad1bab8f4d879647fb34755e4f2bcb" canonical_url: "https://aidr.today/dcc32713?lang=en" title: "Kyutai Releases Voice of Reason: A Speech-Native Model that Solves Spoken Math with Reinforcement Learning" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-09-23T06:33:18.000Z" category: "Research" topics: ["kyutai","text-to-speech","reasoning","rl","open-source"] source_urls: ["https://www.marktechpost.com/2026/09/22/kyutai-releases-voice-of-reason-a-speech-native-model-that-solves-spoken-math-with-reinforcement-learning/"] summary: "Kyutai has released Voice of Reason, 2 open-weight speech-to-speech models built on GLM-4-Voice-9B. Supervised fine-tuning and reinforcement learning lift spoken GSM8K accuracy from 27.3% to 77.1%. There is no transcription step and no text LLM in the loop. Both checkpoints are on Hugging Face and run on a single H100. The post Kyutai Releases Voice of Reason: A Speech-Native Model that Solves Spoken Math with Reinforcement Learning appeared first on MarkTechPost ." --- # Kyutai Releases Voice of Reason: A Speech\-Native Model that Solves Spoken Math with Reinforcement Learning > [Open the canonical story]() **Published:** 2026-09-23T06:33:18.000Z **Category:** Research **Topics:** kyutai, text\-to\-speech, reasoning, rl, open\-source ## Summary Kyutai has released Voice of Reason, 2 open\-weight speech\-to\-speech models built on GLM\-4\-Voice\-9B\. Supervised fine\-tuning and reinforcement learning lift spoken GSM8K accuracy from 27\.3% to 77\.1%\. There is no transcription step and no text LLM in the loop\. Both checkpoints are on Hugging Face and run on a single H100\. The post Kyutai Releases Voice of Reason: A Speech\-Native Model that Solves Spoken Math with Reinforcement Learning appeared first on MarkTechPost \. ## Sources - [Story source]()