Tencent officially announced the launch of Tencent Hy4 on November 4, 2026, marking the transition of its flagship artificial intelligence system into open-weight availability. The model incorporates visual perception for the first time alongside a 1-million-token context window, while adding computer-use capabilities designed to navigate graphical desktop interfaces and command-line environments. Tencent stated that model weights and technical system cards focusing on cybersecurity and biological safeguards will be released publicly over the coming days.
A prominent design priority in Hy4 is model-product co-design for complex enterprise workflows. In office productivity benchmarks on Tencent WorkBuddy, Hy4 recorded an average score of 83.18 and a 96.15 percent task pass rate, ranking first in nine out of twelve evaluated operational scenarios within the Chinese enterprise ecosystem. Integrated with Tencent WeCom, the model continuously indexes multi-project context across communication channels, calendars, and document repositories to monitor approaching deadlines and prepare background briefings for upcoming meetings.
Financial domain optimization was validated through a large-scale blind evaluation conducted in collaboration with Tencent Financial Technology. The study evaluated 579 real-world tasks judged by 213 financial analysts, of whom 153 were external specialists and 70 percent held more than five years of industry experience. Hy4 achieved a leading composite score of 3.04, outperforming DeepSeek-V4.1-Flash-1017 (2.97), Kimi K3.1 Pro (2.68), and GLM-5.3 (2.68). In external testing on the Investment Research Agent Benchmark (iRaB) created by AI firm Rabyte, Hy4 showed a 12.03 percentage point gain over Hy4 Preview.
To reduce ungrounded outputs in regulated sectors, Tencent implemented targeted search integration focused on cross-source verification. In internal testing, Hy4 reduced its overall hallucination rate from 12.95 percent in the preview edition down to 7.2 percent, while errors stemming from unsupported factual assumptions decreased from 23.1 percent to 5.6 percent. For software and game engineering, collaboration with Tencent Interactive Entertainment Group enabled deep integration with Unreal Engine, Unity, and Godot, producing a 19.1 percent benchmark improvement in GameDev and a 9.8 percent increase in GameLogic tasks.
In scientific research, Hy4 was evaluated on TenureTrail, an assessment comprising 123 executable problems across seven scientific disciplines, demonstrating strong performance in physical simulations and systems engineering. The team applied the Hyra 2.0 harness to advance Recursive Self-Improvement (RSI), utilizing Hy4 to optimize pre-training infrastructure for smaller sister architectures. Through this pipeline, Tencent open-sourced two specialized scientific models: HyCell for cellular biology representations and HyPDE for partial differential equations in physics.
Hy4 is available via Tencent Cloud TokenHub, Yuanbao, CodeBuddy, and WeCom Dayuan, alongside external API listings on OpenRouter and Hermes. Despite substantial architecture updates, API pricing remains locked to preview rates at approximately $0.90 per million input tokens, $2.70 per million output tokens, and $0.045 per million cached read tokens. Tencent announced that a higher-throughput Hy4 Flash variant will launch in the near future.