The Radar

Brain encoding model that predicts fMRI responses across video, audio, and text stimuli.

Meta's cracked the code on unified brain encoding across multiple modalities. This isn't just neuroscience research, it's the foundation for building AI that actually understands how humans process information. The implications for brain-computer interfaces and AI alignment are massive.

Real-time multimodal voice model for low-latency audio, video, and tool use.

Google's answer to OpenAI's Advanced Voice Mode, but with proper multimodal support and developer API access. The native streaming architecture means no more clunky pipeline setups. This is how voice AI should work.

State-of-the-art automatic speech recognition model for enterprise speech intelligence.

Cohere's moving beyond text generation into speech recognition territory. This signals the consolidation of AI capabilities under single providers rather than fragmented specialist tools. Enterprise customers get one throat to choke instead of managing multiple ASR vendors.