3 AI stories you missed: Ming-3.0 voice cloning, Amazon TorchNeuron, MiniMax M3.1 Flash
Three more AI stories from October 19, 2026.
Ant Group's inclusionAI open-sourced five Ming-3.0 speech models
MIT license. The smallest, Ming-3.0-tts-0.5B (a ~3GB weights file on Hugging Face), clones a voice from a short reference clip, designs voices from a text description and speaks 7 languages plus Sichuanese and Cantonese.
Sources
Amazon's TorchNeuron runs PyTorch on Trainium with one config change
Amazon says it runs existing PyTorch code on its Trainium AI chips with a single configuration change, no rewrite. It goes open source at PyTorch Conference North America on Oct 20, a direct challenge to Nvidia's CUDA software advantage.
Sources
MiniMax launched M3.1 Flash
It says online inference is nearly 3x faster and usage costs are 70% lower than its predecessor's, at $0.15 per million input tokens and $0.60 per million output tokens. MiniMax says model weights will be released later.
Sources