Benchmark Almanac

Tencent Hy4 preview

Tencent Hy4 preview ranks first on 2 of the 64 benchmarks it has scores on.

Tencent Released 28 Aug 2026
79
Benchmarks with a score
2
Ranked first, out of five or more models
4
In the top three
Tencent Aug 2026

Tencent Hy4 preview

Open weights

No Epoch Capabilities Index score yet
  • NL2Repo-Bench 58.9% #5 /28
  • WideSearch 83.9% #1 /17
  • Harvey LAB (Legal Agent Benchmark) 9.17% #21 /80

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.

Knowledge and language

BenchmarkScoreRankSource
WideSearch 83.9% #1 of 17 Model card
Humanity's Last Exam 55.4% #22 of 147 Model card
Extended NYT Connections 68.2% #55 of 97 Official leaderboard

Math and science

BenchmarkScoreRankSource
GPQA Diamond 92.3% #16 of 284 Model card
ProofBench 75% #16 of 83 Epoch AI
MathArena (Apex, IMO, USAMO) 74.2% #4 of 13 Model card
ProofBench v1.1 75% #17 of 51 Official leaderboard
HorizonMath 8.80% #1 of 3 Model card
CritPt 16.9% #6 of 14 Model card
ExplorationBench 63.3% #7 of 10 Official leaderboard
xbench-ScienceQA 69.8% #21 of 29 Official leaderboard
ArXivMath 66.6% #6 of 8 Model card
BioMysteryBench 71.3% #6 of 8 Model card
BrokenArXiv 54.6% #1 of 1 Model card

Coding

BenchmarkScoreRankSource
SWE-Bench Pro 65.7% #11 of 77 Model card
SWE-bench Multilingual 82.9% #10 of 70 Model card
Code Arena 1,631 #21 of 124 Official leaderboard
NL2Repo-Bench 58.9% #5 of 28 Model card
Code Migration 47.4% #14 of 78 Official leaderboard
Vibe Code Bench 77.5% #34 of 115 Official leaderboard
Hy-SWE Max 64.2% #1 of 3 Model card
SWE Atlas 58.4% #4 of 11 Model card
IOI v2 (Vals) 59.3% #18 of 45 Official leaderboard
Tencent WorkBuddy Bench 76.6% #5 of 10 Official leaderboard
Hy-Backend 2.0 35.2% #1 of 2 Model card
DeepSWE 64.3% #29 of 51 Model card
Code Migration v1.1 45.4% #19 of 29 Official leaderboard
SWE-Marathon 31.9% #9 of 11 Model card
ProgramBench 17.5% #20 of 20 Model card
PostTrainBench 1.1 35.6% #2 of 2 Model card
Hy-CompanyBench V2 62.4% #1 of 1 Model card

Agents and tools

BenchmarkScoreRankSource
MCP Atlas 83.7% #7 of 40 Model card
SkillsBench 62.9% #3 of 13 Model card
Terminal-Bench 2.1 85.4% #14 of 53 Model card
Long-Horizon-Terminal-Bench 2.0 53.2% #5 of 15 Official leaderboard
Agent Arena 0.01 #25 of 53 Official leaderboard
AutomationBench 32.1% #24 of 41 Model card
Toolathlon-Verified 74.1% #8 of 13 Model card
DRACO 77.2% #10 of 11 Model card
Hermes Bench 58.6% #16 of 17 Official leaderboard
Harbor-Index 39.6% #1 of 1 Model card
Hy-BrowseComp-Pro2 56.1% #1 of 1 Model card
Hy-LifeSearch 49.2% #1 of 1 Model card

Professional work

BenchmarkScoreRankSource
$OneMillion-Bench 35.3% #2 of 13 Official leaderboard
Public Benefits Bench 68.6% #9 of 50 Official leaderboard
Legal Research Bench 45.2% #20 of 88 Official leaderboard
Tax Agent Bench 32.6% #18 of 79 Official leaderboard
Harvey LAB (Legal Agent Benchmark) 9.17% #21 of 80 Official leaderboard
Vals Finance Agent v2 55.1% #23 of 80 Official leaderboard
BankerToolBench 78.6% #1 of 3 Model card
GDPval-AA v2 1,678 #9 of 26 Model card
LegalBench 83.8% #50 of 138 Official leaderboard
Vals Legal Index 71.7% #12 of 33 Official leaderboard
Vals Finance Index 57.5% #14 of 35 Official leaderboard
APEX-Agents 37.1% #5 of 12 Model card
MedScribe 83.6% #43 of 98 Official leaderboard
Vals Finance Agent v2.1 61.6% #16 of 35 Official leaderboard
MedCode 43.3% #45 of 96 Official leaderboard
Excel Modeling Benchmark 58.1% #38 of 76 Official leaderboard
JobBench 61.7% #6 of 12 Model card
E-Bench (Tencent) 77.1% #1 of 2 Model card
Excel Modeling Benchmark v1.1 69% #21 of 37 Official leaderboard
OfficeQA Pro 66.2% #8 of 14 Model card
Harvey LAB v1.1 (HLAB-Vals-1.1) 8.75% #22 of 33 Official leaderboard
WorkSpaceBench 60.2% #2 of 3 Model card
Vals Index v2.2 53.2% #22 of 32 Official leaderboard
OR-License 66.5% #8 of 10 Official leaderboard
Agents' Last Exam 22.8% #23 of 25 Model card
E-Bench-Code (Tencent) 79% #1 of 1 Model card
Hy-FinAgentBench 79.7% #1 of 1 Model card
Hy-FinmodelBench v2 57% #1 of 1 Model card

Vision, audio and robotics

BenchmarkScoreRankSource
ScienceVision 0.16 #11 of 11 Official leaderboard

Safety and security

BenchmarkScoreRankSource
SUPERChem 66.4% #1 of 5 Model card
CyberGym 78.4% #19 of 33 Model card
CyberBench v1.1 (Vals) 65.4% #28 of 43 Official leaderboard
SRE-Bench 2.29% #22 of 33 Official leaderboard
CyberBench (Vals) 57.5% #18 of 26 Official leaderboard
CWE-bench 53% #10 of 12 Official leaderboard

Other

BenchmarkScoreRankSource
Hermes Index 26.1 #16 of 17 Official leaderboard

Its launch post

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