Contrastive-LM has released CLM-8B, an open System One model designed for agent action scoring rather than text generation. The model adds two small projection heads to a frozen Qwen3-8B encoder and trains them with a contrastive InfoNCE objective, achieving up to 9× faster inference than comparable systems in zero-shot tests (MarkTechPost). The approach trades text generation for structured scoring of candidate actions against state, reducing inference latency for agent orchestration workflows.
Google released Gemini 3.8 Flash TTS and Flash-Lite TTS, new text-to-speech models available via the Gemini API and Google AI Studio (Google DeepMind). Flash TTS supports over 100 languages and generates custom voices from natural language prompts, ranking first on Hume AI’s Voice Design Benchmark with a score of 71.4 (MarkTechPost). NVIDIA also released Nemotron 3 Diarization, an open-weight 100M-parameter speaker diarization model tracking up to 8 speakers in real time, including overlapping speech in both offline recordings and real-time streams (MarkTechPost).