Google and Cohere released new audio-focused AI models this week, with Google's Gemini 3.1 Flash Live targeting customer service automation and Cohere's unnamed model focused on speech transcription. Both companies claim "significantly higher output quality" than previous versions, but neither provided concrete benchmarks, performance metrics, or detailed technical specifications that developers actually need.
This pattern of vague capability claims without substance is becoming tiresome in the AI space. Audio processing is notoriously difficult to get right — latency, accuracy, accent handling, and noise filtering all matter immensely in production. When OpenAI launched their real-time voice API, they at least provided clear latency numbers and quality samples. Here, we get marketing speak about "optimization" without the data to back it up.
What's particularly frustrating is that my research across Google's own properties turned up nothing but generic Chrome browser pages and search interfaces. No developer documentation, no API endpoints, no pricing — just the usual corporate digital tumbleweeds. For companies supposedly launching new models, the information architecture suggests these aren't ready for serious developer adoption.
If you're building audio applications, wait for actual benchmarks and real-world testing before jumping on these releases. The AI audio space is moving fast, but substance matters more than announcements. Until we see concrete performance data, treat these as placeholder launches rather than production-ready tools.