Voice of Reason takes spoken math problems and answers in speech without an intermediate transcript or text LLM, according to its team. The report gives it 77.1% on GSM8K, compared with 27.3% for GLM-4-Voice.
Voice of Reason was presented as a system that receives spoken math problems, reasons about them, and responds in speech. According to the report, its process uses neither an intermediate transcript nor a text LLM.
On the GSM8K evaluation, the system scored 77.1%; GLM-4-Voice scored 27.3%, according to the reported figures. These results describe the reported comparison and do not guarantee similar performance on other tasks or under different conditions.
When studying or applying this kind of approach with AI, avoid putting personal data or internal information into examples. Use synthetic or anonymized problems, and verify results before relying on them in decisions or products.
Also check what data is sent, how it is handled, and what is logged. Keep public material separate from confidential content and follow your organization’s security rules.