Tencent released Hy4 preview, a 770-billion-parameter language model with 49B active parameters and over 1 million token context window. It's positioned as a general-purpose model for productivity work—coding, office tasks, scientific research—and ranks competitively against other open-source models. The company is offering it free on its WorkBuddy and CodeBuddy platforms for two weeks, with API access available through Tencent Cloud and OpenRouter at $0.834 per million input tokens and $2.501 per million output tokens. In internal benchmarking across 163 experts evaluating 203 engineering tasks, Hy4 preview scored 2.99 out of 4.00, edging out GLM-5.3 (2.92) and Kimi K3 (2.94). The model was trained on high-quality data from Tencent's internal teams in software engineering, gaming, finance, and security. For software development, it handles long-context tasks with better debugging and planning. For office work, it's optimized for financial analysis and cross-document collaboration. In game development, it can generate playable prototypes from text descriptions. For scientific work, it shows improvements in AI research, molecular dynamics, physics, and mathematics. What sets this release apart is that Hy4 preview participated in its own development—the first time Tencent says this happened at scale. The model identified bottlenecks in its inference system and autonomously optimized operators and communication layers, achieving 31.8% throughput improvement over baseline. It also proposed training methods, ran experiments, and iterated on results, creating what Tencent calls an "early-stage recursive self-improvement loop." More models in the Hy4 series are expected soon.