OpenCode Integrates Tencent Hy3 Preview for Low Cost Reasoning and Long Context

OpenCodeOpenCode

OpenCode added Tencent's new Hy3 preview model to its platform, offering free access for two weeks. The model features a 256K context window and a reasoning-focused architecture designed to deliver high performance while reducing inference costs.

OpenCode, an open-source AI coding agent, integrated the Hy3 preview model from Tencent. This text-only reasoning model features a 256K context window (the data a model processes at once) and uses a Mixture-of-Experts architecture (where only specific sub-networks activate per task) with 295B total and 21B active parameters.
Context window
256K tokens
Total parameters
295B
Active parameters
21B
Modality
Text only
Availability
Free for 2 weeks via OpenCode
Architecture
Mixture of Experts

The integration mirrors the launch of Tencent's controllable reasoning model and follows OpenCode's recent addition of Opus 4.7 for massive context. By adding Hy3, OpenCode provides a reasoning-heavy alternative that prioritizes AI economics. The 21B active parameter count enables frontier-level logic without the high token costs of dense models.

Use Hy3 preview within OpenCode to handle complex debugging and refactoring. This update extends the platform's free high-speed agentic coding for developers. The 256K context window is sufficient for most repositories, allowing you to test the model's reasoning capabilities in autonomous workflows before it moves to standard pricing.

View the full update on opencode.ai
OpenCode
OpenCode
@opencode
X

OpenCode x Tencent Hy3 preview — free for 2 weeks Strongest in the Hy series - 256K context · reasoning · text only - 295B (A21B) - ☝️means it's built for strong performance at lower token cost; ur welcome

31retweets751likes
View on X

Still wondering? A few quick answers below.

Tencent Hy3 preview is the latest and most capable model in the Hunyuan series, specifically designed for complex reasoning tasks. It is a text-only model that supports a 256K context window, allowing it to process large amounts of information like entire codebases. It is currently available as a preview through the OpenCode agent platform.

Users can access the Tencent Hy3 preview model for free through the OpenCode platform for a two-week period starting from the announcement date. This promotional window allows developers to test the model's reasoning and long-context capabilities in their coding workflows without incurring API costs before the model eventually transitions to standard pricing.

The Hy3 preview is a large-scale model with 295 billion total parameters. It utilizes a Mixture of Experts architecture where only 21 billion parameters are active during any single request. This design allows the model to maintain high performance and reasoning capabilities while significantly reducing the computational resources and token costs required for inference.

No, the Tencent Hy3 preview is a text-only model. While it is the strongest entry in the Hunyuan series to date and features advanced reasoning capabilities with a 256K context window, it does not currently support multimodal inputs like images, audio, or video. It is optimized specifically for text-based tasks and agentic coding workflows.

The Hy3 model achieves lower token costs through its Mixture of Experts architecture, which features 295 billion total parameters but only 21 billion active parameters. By only activating a fraction of the model for each task, it delivers strong performance and reasoning capabilities with much higher efficiency than traditional dense models of a similar size.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

See all AI news & updates from OpenCode →

Share this update