Hy4 preview

Released2026-08-28
Price per 1M$0.83 in · $2.50 out
Overall score58.7 · rank 41
Context window1,024,000 tokens
ProviderTencent
WeightsOpen weights

Scores

DimensionScoreRankBenchmarks
Overall: Balanced 58.7 41 19
Overall: Coding 57.9 51 16
Overall: Reasoning 59.1 42 4
Overall: Agentic work 57.9 57 14
Overall: Frontend 58.1 38 5
Coding 58.0 46 3
Agents 58.0 54 9
Reasoning 57.1 67 2
Math 60.5 27 1
Knowledge 61.3 19 1
Frontend 58.2 39 2
General 60.3 28 1
Value vs price+10.330

Evidence

Coding 58.0 ±3.3

BenchmarkResultSourceDate
Code Migration v1.1 45.4% Vals AI 2026-11-09
IOI 59.3% Vals AI 2026-10-29
ProgramBench 0.0% Vals AI 2026-11-09

Agents 58.0 ±1.3

BenchmarkResultSourceDate
Terminal-Bench 4.0 (Vals run) 8.1% Vals AI 2026-11-09
SRE Bench 2.3% Vals AI 2026-11-05
Terminal-Bench-Science 1.4% Vals AI 2026-11-09
Terminal-Bench 2.1 (Vals run) 55.1% Vals AI 2026-10-01
EMB v1.1 (Excel financial modeling) 68.9% Vals AI 2026-11-09
Tax Agent Bench 32.6% Vals AI 2026-11-09
Finance Agent v2.1 61.6% Vals AI 2026-11-09
HLAB-Vals 1.1 (legal agent) 8.8% Vals AI 2026-11-09
CyberBench v1.1 65.4% Vals AI 2026-11-05

Reasoning 57.1 ±3.2

BenchmarkResultSourceDate
Extended NYT Connections 68.2% Lech Mazur 2026-11-10
BioMysteryBench 69.3% Vals AI 2026-11-09

Math 60.5 ±2.9

BenchmarkResultSourceDate
ProofBench v1.1 75.0% Vals AI 2026-11-09
ProofBench 75.0% Epoch AI 2026-11-11

Knowledge 61.3 ±6.5

BenchmarkResultSourceDate
Legal Research Bench 45.2% Vals AI 2026-11-09

Frontend 58.2 ±0.7

BenchmarkResultSourceDate
Vibe Code Bench v1.1 77.5% Vals AI 2026-11-09
LMArena WebDev 1,631 LMArena 2026-11-09

General 60.3 ±3.3

BenchmarkResultSourceDate
LLM Debate Benchmark 1,577 Lech Mazur 2026-10-26

Names matched to this model

hy4-preview, tencent-hy4-preview, tencent-hy4-preview-high

Compare Hy4 preview with other models