Technology
Tencent releases and open-sources Hy4 preview large language model
The Chinese tech giant launched a 770B parameter model featuring a 1M+ token context window and self-optimization capabilities.
The short version
- Tencent open-sourced its Hy4 preview large language model, which features 770 billion total parameters, 49 billion active parameters, and a context window exceeding 1 million tokens.
- The model was deployed for public use, offering two weeks of free access on platforms like WorkBuddy and CodeBuddy alongside global API access via Tencent Cloud TokenHub and OpenRouter.
- Tencent reported that Hy4 preview participated directly in its own development and autonomous inference optimization, boosting throughput by 31.8%.
- The company plans to launch additional models in the Hy4 series following this preview release.
Key facts
- Tencent released and open-sourced Tencent Hy4 preview, featuring 770 billion total parameters, 49 billion active parameters, and a context window exceeding 1 million tokens.[Hacker News]
- Hy4 preview is available globally for free for two weeks on WorkBuddy and CodeBuddy, with API pricing set at $0.834 per million input tokens and $2.501 per million output tokens.[Hacker News]
- According to internal blind evaluation scores by Tencent, Hy4 preview scored 2.99 out of 4.00 across 203 engineering tasks, compared to GLM-5.3 at 2.92 and Kimi K3 at 2.94.[Hacker News]
- Tencent states that the model participated in its own training optimization and autonomously optimized its inference system, increasing end-to-end throughput by 31.8% over the baseline.[Hacker News]
What remains uncertain
- The reported benchmark performance and 31.8% inference throughput gain rely solely on internal Tencent evaluations and self-reported metrics.[Hacker News]