Tencent open-sources Hy3, a 295-billion-parameter LLM

Tencent has released the Hy3 LLM (Hyperbolic Model 3) under the permissive Apache 2.0 license. It is one of the few open models from the Chinese AI leader that the company has made available to the community for development and research.

Hy3 specifications

The model uses a Mixture-of-Experts (MoE) architecture and has 295 billion parameters. During inference, however, only 21 billion parameters are active, which lowers compute requirements without sacrificing performance. An MTP layer adds 3.8 billion parameters for tuning outputs. The context window is 256K tokens, enough to analyze long documents, codebases or conversations.

The full model takes up 598 gigabytes on disk. The quantized FP8 version takes 300 gigabytes, which makes it easier to deploy on standard hardware.

Performance

Tencent claims that Hy3 outperforms open models with a similar parameter count in benchmarks and delivers results comparable to flagship models that are 2–5 times larger. The company gathered feedback from 50+ products and app developers and refined the model after the Preview version released in April 2026.

• Total parameters: 295 billion
• Active parameters at inference: 21 billion
• Context window: 256K tokens
• Full model size: 598 GB
• Quantized (FP8) size: 300 GB
• License: Apache 2.0

Availability and deployment

The model is hosted on Hugging Face and available for free on the OpenRouter platform until July 21, 2026. This lets developers test the model before deploying it on their own infrastructure. Apache 2.0 means companies can use Hy3 in commercial products without restrictions, provided they comply with the license terms (mainly attribution).

Why it matters

The Hy3 release strengthens Tencent's position as a significant player in the global LLM market. Until now, most open and competitive models came from Western labs (Meta, Mistral, OpenAI) or from other players (Alibaba and ByteDance in China). With its 256K context and MoE architecture, Hy3 shows that Chinese developers are reaching parity in model scale and capabilities.

For the developer community, this widens the choice of open models and allows more flexibility in picking tools for specific tasks without being tied to closed APIs.

What this means

The global open LLM market is becoming more competitive, which pushes down API and infrastructure prices and speeds up innovation. Companies like Tencent invest in open models as a strategy to win community trust and strengthen their AI brand. Free access on OpenRouter until mid-July is a marketing move that lets the company quickly attract users and collect feedback.