Nov 4, 2026
Tencent's Hy4 evaluated on the Vals Index
We evaluated Tencent’s new open-weight Hy4 on Vals Index v2.2 and its components.
-
#13 of 33 on the Vals Index (59.35%) at $1.72 per test, the #4 open-weight model behind Kimi K3.1 Pro (62.52%), DeepSeek V4.1 Flash 1017 (61.26%) and GLM 5.3 (59.73%), and up from 53.21% for Hy4 Preview.
-
#2 open-weight model on Terminal-Bench 4.0 (39.39%, #15 of 60), up from 8.08% for Hy4 Preview. It is also the #3 open-weight model on Finance Agent v2.1 (64.70%, #11 of 36) and Tax Agent Bench (34.40%, #15 of 81), and scores 81.01% on Vibe Code Bench (#29 of 124).
-
Weaker on EMB v1.1 (66.18%, #28 of 38), Legal Research Bench (41.35%, #28 of 90) and HLAB-Vals-1.1 (12.50%, #16 of 34).
Hy4 is a reasoning model with a 1M-token context window and 64k max output tokens, priced at $0.83 per million input tokens and $2.50 per million output tokens. We accessed it through OpenRouter at temperature 0.9 and top_p 1.0; costs use Tencent’s list price.
Congrats to the Tencent team on the release!