All tags
Model: "gpt-oss-puzzle-88b"
not much happened today
voxtral-tts gpt-oss-puzzle-88b gpt-oss-120b elevenlabs-flash-v2.5 mistral-ai elevenlabs nvidia intel huggingface openai text-to-speech model-optimization mixture-of-experts transformer gpu throughput latency multilinguality model-efficiency
Mistral AI announced the release of Voxtral TTS, a 3-billion-parameter text-to-speech model with open weights that outperforms ElevenLabs Flash v2.5 in human preference tests. It runs efficiently on about 3 GB RAM, achieves 90-millisecond time-to-first-audio, and supports nine languages. Meanwhile, NVIDIA released gpt-oss-puzzle-88B, an 88-billion-parameter Mixture-of-Experts Transformer optimized for NVIDIA H100 hardware, improving throughput by 1.63× in long-context scenarios and maintaining or exceeding accuracy of its gpt-oss-120b parent model. Intel plans to launch a new GPU with 32GB VRAM priced at $949, offering 608 GB/s bandwidth and 290W power consumption, targeting local AI workloads.