Benchmark Almanac

Tencent Hy4 preview

Tencent Hy4 preview ranks first on 5 of the 93 benchmarks with five or more models scored.

Tencent Released August 28th, 2026 Compare with others
106
Benchmarks with a score
5
Ranked first, out of five or more models
8
In the top three
Tencent Aug 2026

Tencent Hy4 preview

Open weights

159.3 #36 of 404
Self-Reported Index, from the lab's own numbers. No Epoch Capabilities Index score yet.
  • NL2Repo-Bench 58.9% #7 /30
  • WideSearch 83.9% #1 /19
  • GDP.pdf 35% #17 /19

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 19 Lab model card
Humanity's Last Exam 55.4% #24 of 154 Lab model card
StudyArena Text 1,229 #46 of 85 Official leaderboard
Hemingway-bench 1,014 #23 of 42 Official leaderboard
Extended NYT Connections 68.2% #61 of 103 Official leaderboard
ComplexConstraints 49.2% #12 of 15 Official leaderboard
HANDBOOK.md 24.3% #13 of 16 Official leaderboard

Math and science

BenchmarkScoreRankSource
GPQA Diamond 92.3% #20 of 302 Lab model card
MathArena (Apex, IMO, USAMO) 74.2% #1 of 12 Lab model card
HorizonMath 8.80% #1 of 3 Lab model card
ProofBench v1.1 75% #18 of 52 Official leaderboard
CritPt 16.9% #5 of 13 Lab model card
BioMysteryBench 69.3% #16 of 30 Official leaderboard
ExplorationBench 63.3% #7 of 10 Official leaderboard
xbench-ScienceQA 69.8% #21 of 29 Official leaderboard
ArXivMath 66.6% #7 of 9 Lab model card
Riemann-bench 24% #18 of 20 Official leaderboard
BrokenArXiv 54.6% #1 of 1 Lab model card

Coding

BenchmarkScoreRankSource
SWE-Bench Pro 65.7% #12 of 81 Lab model card
SWE-bench Multilingual 82.9% #11 of 73 Lab model card
Code Migration 47.4% #14 of 78 Official leaderboard
Code Arena 1,632 #23 of 128 Official leaderboard
NL2Repo-Bench 58.9% #7 of 30 Lab model card
Vibe Code Bench 77.5% #39 of 120 Official leaderboard
Hy-SWE Max 64.2% #1 of 3 Lab model card
SWE Atlas 58.4% #5 of 14 Lab model card
IOI v2 (Vals) 59.3% #18 of 45 Official leaderboard
Tencent WorkBuddy Bench 76.6% #6 of 14 Official leaderboard
Hy-Backend 2.0 35.2% #1 of 2 Lab model card
Doc2Harness 38.9% #10 of 19 Official leaderboard
DeepSWE 64.3% #33 of 56 Lab model card
Real-SWE 1.1 33.6% #9 of 15 Official leaderboard
sudo L7 15.6% #11 of 18 Official leaderboard
Code Migration v1.1 45.4% #20 of 30 Official leaderboard
DAYJOB: Data Science 3.30% #12 of 15 Official leaderboard
OpenDesign Arena 74.6 #17 of 21 Official leaderboard
SWE-Marathon 31.9% #9 of 11 Lab model card
PostTrainBench 1.1 35.6% #2 of 2 Lab model card
Hy-CompanyBench V2 62.4% #1 of 1 Lab model card

Agents and tools

BenchmarkScoreRankSource
MCP Atlas 83.7% #9 of 46 Lab model card
Terminal-Bench 2.1 85.4% #16 of 56 Lab model card
SkillsBench 62.9% #4 of 14 Lab model card
RuneBench 5.37 #25 of 76 Official leaderboard
Long-Horizon-Terminal-Bench 2.0 53.2% #5 of 15 Official leaderboard
Toolathlon-Verified 74.1% #11 of 21 Lab model card
Agent Arena 0.47% #30 of 56 Official leaderboard
AutomationBench 32.1% #30 of 47 Lab model card
DuelLab GameBench 2 60.8 #37 of 57 Official leaderboard
DRACO 77.2% #11 of 12 Lab model card
Hermes Bench 58.6% #16 of 17 Official leaderboard
Harbor-Index 39.6% #1 of 1 Lab model card
Hy-BrowseComp-Pro2 56.1% #1 of 1 Lab model card
Hy-LifeSearch 49.2% #1 of 1 Lab model card

Professional work

BenchmarkScoreRankSource
$OneMillion-Bench 35.3% #2 of 13 Official leaderboard
Public Benefits Bench 68.6% #9 of 50 Official leaderboard
EnterpriseBench: CoreCraft 64.6% #10 of 51 Official leaderboard
Legal Research Bench 45.2% #24 of 93 Official leaderboard
Tax Agent Bench 32.6% #22 of 84 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
FoodTruck Bench $39,037 #17 of 59 Official leaderboard
GDPval-AA v2 1,678 #9 of 27 Lab model card
BankerToolBench 78.6% #1 of 3 Lab model card
LegalBench 83.8% #50 of 138 Official leaderboard
E-Bench (Tencent) 77.1% #3 of 8 Official leaderboard
E-Bench-Code (Tencent) 79% #3 of 8 Official leaderboard
DAYJOB: Healthcare 6.40% #14 of 37 Official leaderboard
Vals Legal Index 71.7% #15 of 38 Official leaderboard
Vals Index v2 55.4% #27 of 65 Official leaderboard
APEX-Agents 37.1% #5 of 12 Lab model card
Tuesday Work Index 53.8 #19 of 44 Official leaderboard
MedScribe 83.6% #43 of 97 Official leaderboard
DAYJOB: Finance 9.30% #18 of 39 Official leaderboard
MedCode 43.3% #45 of 96 Official leaderboard
Vals Finance Index 57.5% #19 of 40 Official leaderboard
Excel Modeling Benchmark 58.1% #38 of 76 Official leaderboard
Vals Finance Agent v2.1 61.6% #21 of 40 Official leaderboard
GDP.xlsx 18.3% #11 of 20 Official leaderboard
OfficeQA Pro 66.2% #9 of 16 Lab model card
Excel Modeling Benchmark v1.1 69% #25 of 42 Official leaderboard
DAYJOB: Tax 1.60% #9 of 15 Official leaderboard
JobBench 61.7% #9 of 15 Lab model card
Vals Index v2.1 49.9% #31 of 49 Official leaderboard
DAYJOB: Accounting 8.40% #11 of 17 Official leaderboard
WorkSpaceBench 60.2% #2 of 3 Lab model card
Harvey LAB v1.1 (HLAB-Vals-1.1) 8.75% #26 of 38 Official leaderboard
DAYJOB: Law 9.90% #11 of 16 Official leaderboard
OR-License 66.5% #8 of 11 Official leaderboard
Vals Index v2.2 53.2% #27 of 37 Official leaderboard
Agents' Last Exam 22.8% #24 of 26 Lab model card
Hy-FinAgentBench 79.7% #1 of 1 Lab model card
Hy-FinmodelBench v2 57% #1 of 1 Lab model card

Vision, audio and robotics

BenchmarkScoreRankSource
GDP.pdf 35% #17 of 19 Official leaderboard
ScienceVision 0.16 #12 of 12 Official leaderboard

Safety and security

BenchmarkScoreRankSource
AILuminate Safety 1.2 96.0 #1 of 5 Official leaderboard
MLCommons AILuminate 96% #1 of 5 Official leaderboard
SUPERChem 66.4% #1 of 5 Lab model card
nocot-bench (NCRI) 82.8 #51 of 185 Official leaderboard
CyberGym 78.4% #20 of 34 Lab model card
SRE-Bench 2.29% #23 of 35 Official leaderboard
CyberBench v1.1 (Vals) 65.4% #29 of 44 Official leaderboard
CyberBench (Vals) 57.5% #18 of 26 Official leaderboard
CWE-bench 2 47.3% #10 of 14 Official leaderboard
CWE-bench 53% #10 of 12 Official leaderboard

Other

BenchmarkScoreRankSource
SuperCLUE 43.0 #9 of 16 Official leaderboard
Hermes Index 26.1 #16 of 17 Official leaderboard

Its lab's numbers put it at 159.3, number 36 of 404

The Self-Reported Index fits one score to the 23 capability results Tencent reported for this model, on the scale of the Epoch Capabilities Index. Labs choose which benchmarks to show and how to run them, so read it as the lab's claim, not an independent measurement. The 90% interval runs from 157.2 to 161.5. How we compute it.

Its lab's numbers next to 4 leaderboards

Where the benchmark's own leaderboard, or Epoch AI's run, also has this model. Gaps mostly come from different harnesses, effort levels and task subsets. Every pair.

Benchmark Lab says Board says Gap
$OneMillion-Bench 65.4% OneMillionBench (with tools) 35.3% Leaderboard +30 pts
BioMysteryBench 71.3% BioMysteryBench 69.3% Leaderboard, Vals AI harness, v11 full set +2.0 pts
E-Bench-Code (Tencent) 79% E-Bench-Code (Internal) 79% Leaderboard, highest available reasoning Same
E-Bench (Tencent) 77.1% E-Bench (Internal) 77.1% Leaderboard, highest available reasoning Same

Its launch post

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

Hy4 preview model card (Hugging Face) Aug 28, 2026