夜雨聆风minimax-2.5qwen3-coder-nextqwen3.5-a35b-3bnemotron-supernemotron-nano
MiniMax-M2.5 is an MoE model (230B total) extensively trained with reinforcement learning in hundreds of thousands of real-world environments, delivering SOTA results in coding, agentic tool use, search, and office work.
Qwen3.5 represents a significant leap forward, integrating breakthroughs in multimodal learning, architectural efficiency, reinforcement learning scale, and global accessibility. It has 35B total parameters and 3B activated, supporting a native context length of 262,144 tokens.
Qwen Coder Next is an 80B MoE with 3B active parameters designed for coding agents and local development. Excels at long-horizon reasoning, complex tool usage, and recovery from execution failures.
Nemotron 3 Nano by NVIDIA
General purpose reasoning and chat model trained from scratch by NVIDIA. Contains 30B total parameters with only 3.5B active at a time for low-latency MoE inference.
Features a reasoning toggle to enable or disable intermediate reasoning traces, with improved accuracy on complex queries when reasoning is enabled. Includes native agentic capabilities for tool use, making it suitable for AI agents, RAG systems, chatbots, and other AI-powered applications.
Supports a context length of 1M tokens.
Nemotron 3 Super
General purpose reasoning and chat model trained by NVIDIA. Contains 120B total parameters with only 12B active at a time, using a hybrid LatentMoE architecture with Multi-Token Prediction layers for efficient high-throughput inference.
Supports multiple languages including English, Spanish, French, German, Japanese, Italian, and Chinese.
Supports a context length of 1M tokens.
