Tencent releases Hy4 — multimodal LLM with vision and computer control

🤖 Tencent releases Hy4 — LLM with vision and PC control

Tencent released Hy4 on November 4, 2026 — the first flagship with image understanding and computer control: CLI, GUI, mouse, windows, and training on Blender and CAD. Context is 1M tokens, output up to 65,536. According to Tencent's figures: E-Bench 2.0 — 70 vs. 65.3 for GPT-6.1 Sol, Terminal-Bench 4.0 — 44.4 vs. 55.5.

🌍 Hy4 enters the computer use and agentic tasks segment: APEX-Agents — 75.2%, OSWorld 2.1 — 78.9%, but there is a lag in the terminal — 'controlling a computer' and 'working in the console' are currently different skills.

👤 Hy4 is already available via Tencent Cloud API and OpenRouter (tencent/hy4): 50% discount, input $0.834 and output $2.501 per 1M tokens. Weights are promised later, and Hy4 Flash is in preparation. Computer use may transmit screen content to servers.

Source 1: https://www.tencent.com/tencent-launches-hy4-joining-the-ranks-of-the-worlds-leading-open-source-ai-models/ Source 2: https://openrouter.ai/tencent/hy4