Nobody independent has posted scores for Hy4 on Epoch AI's Capabilities Index or the other tests on this scoreboard². Tencent's Hy team dated the release Nov. 4, 2026¹. OpenRouter lists $0.42 per million input tokens and $1.25 per million output tokens, with a 1,048,576-token context window³. Tencent says the model is on Tencent Cloud and OpenRouter and in products including CodeBuddy, WorkBuddy, Yuanbao and ima¹.

Tencent says Hy4 scales pre-training and post-training relative to Hy4 preview, adds vision for the first time, and improves instruction following, long-context understanding, long-horizon tasks and deep research¹. Tencent says it can read fine-grained image detail and charts, and that computer use covers command-line and graphical interfaces, including software such as Blender and CAD¹. Tencent says GameDev is 19.1% higher and GameLogic 9.8% higher than Hy4 preview¹. Those percentages are Tencent's; they are not Epoch, ARC Prize, Artificial Analysis or Mercor runs^1,2.

The field already ranked on this board is different. Claude Fable 5.2 ranks first of 44 on Epoch AI's Capabilities Index at 174.2; GPT-6 Astra is second at 169.2; Google DeepMind's Gemini 4 Argon is seventh at 166.5². Hy4 has no Epoch rank and no previous-from-same-lab row on the scoreboard².

What is not known is any independent GPQA Diamond, FrontierMath, ARC-AGI-2, SciCode, APEX-Agents or SimpleQA Verified number, and whether Tencent's GameDev and GameLogic figures would hold under an outside harness^1,2. This panel does not assign Hy4 a rank among the 44 models already on Epoch AI's index².