Tencent Hy3

Tencent Hy3 ranks first on 3 of the 43 benchmarks it has scores on.

Tencent Released 2 Jul 2026 Family Hy3 The lab's model card

Benchmarks with a score

Ranked first, out of five or more models

In the top three

Tencent Jul 2026

Open weights Flagship

No Epoch Capabilities Index score yet

Where it ranks

Its best score on each benchmark, and where that ranks among every model with a score there. We rank each model once, at its best setting.

Benchmark	Score	Rank	Source
AA-LCR (Artificial Analysis Long Context Reasoning)	73.4%	#4 of 31	Model card
CL-bench	23.8%	#1 of 6	Model card
LMArena (Elo)	1,454	#58 of 333	Official leaderboard
Humanity's Last Exam	53.2%	#29 of 147	Model card
WideSearch	76.4%	#5 of 17	Model card
CL-bench Life	17%	#1 of 2	Model card
Continual Learning Exam	27.3%	#7 of 10	Official leaderboard
Extended NYT Connections	41.2%	#69 of 97	Official leaderboard

Benchmark	Score	Rank	Source
GPQA Diamond	90.4%	#28 of 284	Model card
IMO-AnswerBench	90%	#4 of 37	Model card
PHYBench	77.4%	#1 of 5	Model card
DTBench	82.2%	#53 of 178	Official leaderboard
MathArena (Apex, IMO, USAMO)	38.7%	#8 of 13	Model card
HorizonMath	7.10%	#2 of 3	Model card
xbench-ScienceQA	66.6%	#24 of 29	Official leaderboard
USAMO 2026	72%	#5 of 6	Model card
Prophet Arena	84.4	#17 of 18	Official leaderboard
PaperBenchX	2.15%	#10 of 10	Official leaderboard
ArXivMath	52.2%	#8 of 8	Model card
FrontierScience	48.1%	#7 of 7	Model card
CMT-Benchmark	37.8%	#2 of 2	Model card
Hy-Euler pro	24.2%	#1 of 1	Model card
Hy-Math	60.9%	#1 of 1	Model card

Benchmark	Score	Rank	Source
SWE-bench Verified	78%	#27 of 150	Model card
StructEval	76.9%	#8 of 28	Official leaderboard
SWE-bench Multilingual	75.8%	#29 of 70	Model card
SWE-Bench Pro	57.9%	#32 of 77	Model card
Code Arena	1,505	#56 of 124	Official leaderboard
NL2Repo-Bench	45.6%	#16 of 28	Model card
Hy-SWE Max	49%	#2 of 3	Model card
Tencent WorkBuddy Bench	70.9%	#9 of 10	Official leaderboard
DeepSWE	28%	#50 of 51	Model card
Hy-Backend 2.0	25%	#2 of 2	Model card
Hy-CompanyBench	41.7%	#1 of 1	Model card

Benchmark	Score	Rank	Source
Claw-Eval Pass^3	68.5%	#1 of 11	Model card
BrowseComp	84.2%	#19 of 77	Model card
DeepSearchQA	91%	#7 of 22	Model card
MCP Atlas	79.1%	#13 of 40	Model card
SkillsBench	55.3%	#7 of 13	Model card
WildClawBench	53.6%	#3 of 5	Model card
Toolathlon	48.5%	#23 of 37	Model card
Terminal-Bench 2.1	71.7%	#33 of 53	Model card
Toolathlon-Verified	56.8%	#16 of 25	Official leaderboard
Agent Arena	-0.07	#42 of 53	Official leaderboard
Hy-SkillsWorld	45.8%	#1 of 1	Model card
ProdBench (Tencent)	23%	#1 of 1	Model card

Benchmark	Score	Rank	Source
APEX-Agents	25.6%	#10 of 12	Model card
$OneMillion-Bench	13%	#13 of 13	Official leaderboard
OR-License	45.2%	#10 of 10	Official leaderboard
E-Bench (Tencent)	50.2%	#2 of 2	Model card
Hy-FinModelBench	69%	#1 of 1	Model card

Benchmark	Score		Rank	Source
SUPERChem	54.9%		#4 of 5	Model card

Benchmark	Score	Rank	Source
Conceptual Reasoning Index	52.2	#60 of 149	Official leaderboard
ACCoRD	66.4	#63 of 150	Official leaderboard
SuperCLUE	62.1	#13 of 18	Official leaderboard

The benchmarks the lab chose to headline, with the scores it gave.