Tencent releases HY 2.0 with Think and Instruct versions

Background

HY 2.0 was described by TechNode as Tencent's next-generation AI model, and it was released with a Think version and an Instruct version.^[1] Next-generation here means the newer model in Tencent's Hunyuan line, which the release reports describe as an upgrade.

What happened

Tencent released Hunyuan 2.0, also written HY 2.0, on 5 December 2025.^[1] It came in two versions, Think and Instruct.^[1] The model is a mixture-of-experts (MoE) design with 406 billion total parameters and 32 billion active parameters, and it has a 256K-token context window.^[1]

Key details

Tencent cited major improvements in pre-training data and reinforcement learning for the Think version.^[1] Pre-training data is the text a model learns from before any specialised tuning. Reinforcement learning adjusts a model based on feedback about its outputs.

The model was already integrated into Yuanbao and ima when it was announced, and it was offered through Tencent Cloud APIs.^[1] This meant users could reach the model inside Tencent's apps, and developers could call it through a cloud interface. The parameter count is the number of learned values inside a model. More parameters generally give a model more capacity, but they also mean more memory and computing power are needed to run it.

Why it matters

The release combined a model built into Tencent's own products with a model offered to developers through a cloud service.^[1] Dataconomy reported that Tencent planned to open-source related technologies and models, which is the part of the announcement that matters most to researchers who follow open models.^[2] Open-sourcing a model means publishing its weights and related materials so that others can download, examine and run it.

For a reader new to AI, the release shows how one flagship model can reach people in several forms: a Think version and an Instruct version, built-in features in apps, and access through a cloud API.