MiniMax M2.1 & GLM 4.7 Redefine AI Coding as Soprano-80M Slashes TTS Latency

New AI Models & Releases

MiniMax M2.1 Officially Launched: SOTA Agentic Coding at 10% the Price of Claude Sonnet 4.5
MiniMax unveiled M2.1, a cutting-edge AI model optimized for agentic workflows and engineering tasks, surpassing Claude Sonnet 4.5 and Gemini 3 Pro in benchmarks (72.5% on SWE-bench Multilingual). The model supports multiple programming languages and offers cost savings ($0.30/million input tokens), with a full open-source release scheduled for December 25th.

• MiniMax M2.1 Officially Launched: SOTA Agentic Coding at 10% the Price of Claude Sonnet 4.5
  • MiniMax Official Announcement
  • Text Generation Docs
  • Agent Demo

GLM 4.7 Released on Hugging Face
The GLM 4.7 model is now available on Hugging Face, excelling in web development tasks and diagram generation during reasoning/planning. Early user feedback highlights its versatility for technical applications.

• GLM 4.7 is out on HF!
  • Hugging Face Model Page

Open-Source Tools & Frameworks

Soprano-80M: Ultra-Low-Latency TTS Model for Voice Chatbots
Soprano-80M is a new open-source TTS model achieving <15ms latency and 2000x real-time speed with <1GB VRAM, using a vocoder-based decoder and neural audio codec. Released under Apache 2.0, it’s ideal for real-time voice applications.

• I made Soprano-80M: Stream ultra-realistic TTS in <15ms, up to 2000x realtime, and <1 GB VRAM
  • GitHub Repository
  • Hugging Face Demo
  • Model Weights