Key Info

Tencent Hy4 scores 35 on the Artificial Analysis Intelligence Index v5, equal to Grok 4.7 (xhigh) and the #2 model from a China-based AI lab, at roughly $0.78 per index task.

Highlights

  • Hy4 leads Chinese labs on the FrontierCode 2.0 benchmark, with agentic coding cited as the area where the gap is closing fastest.
  • The model shows an 89% hallucination rate on AA-Omniscience, highlighting a remaining reliability cost.
  • Cost-capability positioning is described as a Pareto shift versus comparable Western models.
  • Hy4 is the successor to Hy3, released earlier this year.