What did Tencent's new Hy4 AI model promise builders this week?

Tencent called Hy4 a full release on November 4, yet the model weights and system cards are still due "in the coming days". A cheaper Hy4 Flash is also on the way.
The promise to builders is a big open model that can now see and act. Hy4 reads images and charts. It can also operate command-line and graphical interfaces, which matters if you want an agent to drive software. Tencent says the hallucination rate dropped from 12.95% to 7.2%, though that number comes from its own tests.
It is a mixture-of-experts model with 770 billion parameters in total and 49 billion active at a time. The context window is 1 million tokens. The August preview went out under Apache 2.0, but running it needs roughly 1.5 TB of VRAM. API prices stay where they were, about $0.90 per million input tokens and $2.70 per million output tokens.
Sources: daily.dev, gigazine.net