DeepSeek V4 Flash 0731 vs Hy4 Preview: Benchmark Comparison

DeepSeek V4 Flash 0731 has the higher score on 3 of 18 shared benchmarks; Hy4 Preview leads on 14.

The largest observed score gap is 26.61 pts on IOI , where Hy4 Preview leads.

Reported ±1 standard-error ranges overlap on 10 of 18 shared benchmarks with comparable uncertainty data. This is not a pairwise statistical significance test.

Shared benchmark results

Scores and reported standard errors come from the latest visible Vals benchmark version available for both models.

Benchmark DeepSeek V4 Flash 0731 Hy4 Preview Gap Reported uncertainty
Vals Index 47.96% ±1.24 49.94% ±1.15 1.98 pts Reported ±1 SE ranges overlap
Harvey's Legal Agent Benchmark 8.33% ±2.15 9.17% ±2.00 0.83 pts Reported ±1 SE ranges overlap
Legal Research Bench 30.29% ±3.19 45.19% ±3.46 14.90 pts Reported ±1 SE ranges do not overlap
LegalBench 77.71% ±0.50 83.76% ±0.41 6.04 pts Reported ±1 SE ranges do not overlap
EMB 56.98% ±3.39 58.08% ±3.14 1.10 pts Reported ±1 SE ranges overlap
Finance Agent (v2) 49.52% ±0.47 55.06% ±0.31 5.55 pts Reported ±1 SE ranges do not overlap
Tax Agent Bench 62.56% ±3.14 63.71% ±3.26 1.15 pts Reported ±1 SE ranges overlap
MedCode 41.41% ±2.15 43.25% ±2.13 1.83 pts Reported ±1 SE ranges overlap
MedScribe 80.36% ±1.97 83.60% ±2.06 3.24 pts Reported ±1 SE ranges overlap
ProofBench v1.1 56.00% ±4.99 75.00% ±4.35 19.00 pts Reported ±1 SE ranges do not overlap
BioMysteryBench 64.44% ±1.28 69.26% ±2.59 4.81 pts Reported ±1 SE ranges do not overlap
Terminal-Bench Science 4.29% ±2.44 1.43% ±1.43 2.86 pts Reported ±1 SE ranges overlap
Code Migration 38.63% ±4.38 47.43% ±4.27 8.80 pts Reported ±1 SE ranges do not overlap
IOI 32.72% ±2.35 59.33% ±4.60 26.61 pts Reported ±1 SE ranges do not overlap
ProgramBench 0.00% ±0.00 0.00% ±0.00 0.00 pts Reported ±1 SE ranges overlap
Terminal-Bench 4.0 18.69% ±3.54 8.08% ±1.34 10.61 pts Reported ±1 SE ranges do not overlap
Vibe Code Bench v1.1 74.74% ±3.44 77.48% ±4.04 2.74 pts Reported ±1 SE ranges overlap
CyberBench v1.1 72.08% ±5.41 65.36% ±5.55 6.73 pts Reported ±1 SE ranges overlap

Performance by category

Category DeepSeek V4 Flash 0731 average Hy4 Preview average
Legal 38.78% 46.04%
Finance 56.35% 58.95%
Healthcare 60.89% 63.42%
Math 56.00% 75.00%
Science 34.37% 35.34%
Coding 32.96% 38.47%
Cyber 72.08% 65.36%

Cost and latency

Cost per test appears only where the benchmark reports it for both models. Latency is the measured completion time for that benchmark.

Benchmark DeepSeek V4 Flash 0731 cost Hy4 Preview cost DeepSeek V4 Flash 0731 latency Hy4 Preview latency
Vals Index $0.85 $1.44 43m06s 1h01m
Harvey's Legal Agent Benchmark $0.06 $0.98 16m51s 29m25s
Legal Research Bench $0.28 $0.85 21m07s 1h04m
LegalBench N/A N/A 3.95s 90.56s
EMB $0.38 $0.79 26m14s 30m39s
Finance Agent (v2) $0.28 $0.59 8m10s 23m45s
Tax Agent Bench $0.16 $0.59 5m37s 41m04s
MedCode N/A N/A 2m07s 6m39s
MedScribe N/A N/A 77.55s 5m50s
ProofBench v1.1 $0.04 $0.34 15m34s 22m54s
BioMysteryBench $0.04 $0.36 27m03s 23m30s
Terminal-Bench Science $1.10 $2.14 1h34m 3h48m
Code Migration $3.61 $3.41 2h08m 2h13m
IOI $0.37 $1.35 27m43s 1h01m
ProgramBench $1.29 $12.81 5h39m 2h10m
Terminal-Bench 4.0 $0.77 $2.10 1h46m 2h10m
Vibe Code Bench v1.1 $0.91 $2.18 37m22s 41m33s
CyberBench v1.1 $0.19 $0.49 15m28s 36m43s

Results available only for DeepSeek V4 Flash 0731

  • TaxEval v2
  • GPQA Diamond
  • MMLU Pro
  • LiveCodeBench
  • SkillsBench
  • SWE-bench

Results available only for Hy4 Preview

  • SRE Bench
  • Public Benefits Bench v1.1
Model details DeepSeek V4 Flash 0731 Model details Hy4 Preview