Hy4 preview benchmark grid
THIS IS NOT A DRILL... TENCENT JUST SHIPPED A 1M CONTEXT MONSTER 🤯 The engineering team at TencentHunyuan just dropped the Hy4 preview. The scale of this model is huge, and the benchmarks back it up. → 770B total parameters → 49B active parameters → 1M token context window for massive codebase retention → Sustained long-horizon execution → Advanced multi-source tool use They built this specifically to handle long-horizon software engineering and multi-step agentic workflows without losing context. let's dive in 🧵↓
文字记录 (en)
Thank you.
