Tencent Launches Hy4 Preview Open-Weight AI Model at Aggressive API Pricing

Tencent has introduced Hy4 Preview, a large open-weight artificial intelligence model that the company is pitching as its most substantial generational leap to date. Unlike closed proprietary systems such as GPT-5.6 and Claude Opus 5, the model's weights are publicly available for developers to download and deploy on their own infrastructure, making it a direct competitor to offerings from DeepSeek, Zhipu AI, Moonshot AI, OpenAI and Anthropic.

The pricing is the headline for cost-conscious developers. Tencent lists API access starting at $0.042 per million cached input tokens, with standard input at $0.834 and output at $2.501 per million tokens. The model is available through Tencent Cloud, the OpenRouter marketplace and Tencent's own AI applications, with weights hosted on supported open-source repositories. It was released on August 28, 2026.

Architecturally, Hy4 Preview is built on a Mixture-of-Experts design that activates only a fraction of its parameters for each token. It carries 770 billion total parameters and activates 49 billion per token, spread across 78 layers. Of those, 77 are MoE layers containing 256 routed experts and one shared expert each. The model also incorporates Gated DeepSeek Sparse Attention and identity Hyper-Connections to improve information processing, and it supports a context window of up to one million tokens.

Tencent says the model brings meaningful gains in software engineering, office productivity, game development and scientific research. According to benchmarks published by the company, Hy4 Preview outperforms several frontier competitors in selected coding and productivity tests, though it acknowledges that Claude Opus 5, GPT-5.6 Sol and some other rivals remain ahead in certain areas. The evaluation covered software engineering, agent-based and reasoning benchmarks, comparing the model against Claude Opus 5, GPT-5.6 Sol, DeepSeek V4 Pro, GLM 5.3 and Kimi K3.

The release reflects a widening trend in which Chinese technology firms compete on openness and price as much as raw capability. By expanding model size, context length and training data across both pre-training and post-training, Tencent is betting that developers will favour a system they can inspect, self-host and run cheaply. For businesses evaluating AI infrastructure, the combination of open weights and low token costs could lower the barrier to building and scaling custom applications.

Tencent describes Hy4 Preview as its largest improvement between model generations so far, a claim that will be tested as independent developers put the system through real-world workloads. The model's performance in production settings, particularly for coding agents and long-document tasks enabled by its extended context, is likely to determine whether the aggressive pricing translates into broader adoption.