DeepSeek V4 vs GLM 5.3 Flash: Benchmark Comparison

DeepSeek V4 has the higher score on 5 of 14 shared benchmarks; GLM 5.3 Flash leads on 8.

The largest observed score gap is 26.26 pts on Terminal-Bench 4.0 , where GLM 5.3 Flash leads.

Reported ±1 standard-error ranges overlap on 4 of 14 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 GLM 5.3 Flash Gap Reported uncertainty
Legal Research Bench 23.08% ±2.93 45.19% ±3.46 22.11 pts Reported ±1 SE ranges do not overlap
LegalBench 80.32% ±0.47 83.93% ±0.43 3.61 pts Reported ±1 SE ranges do not overlap
Tax Agent Bench 25.17% ±3.06 31.55% ±3.35 6.38 pts Reported ±1 SE ranges overlap
TaxEval v2 72.08% ±0.88 75.59% ±0.84 3.52 pts Reported ±1 SE ranges do not overlap
MedScribe 75.14% ±2.00 88.94% ±1.91 13.79 pts Reported ±1 SE ranges do not overlap
ProofBench v1.1 16.00% ±3.69 21.00% ±4.09 5.00 pts Reported ±1 SE ranges overlap
GPQA Diamond 89.39% ±1.65 86.36% ±2.13 3.03 pts Reported ±1 SE ranges overlap
MMLU Pro 87.25% ±0.34 86.06% ±0.34 1.19 pts Reported ±1 SE ranges do not overlap
LiveCodeBench 87.48% ±0.95 80.51% ±1.05 6.97 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
SkillsBench 51.27% ±4.60 40.16% ±4.43 11.12 pts Reported ±1 SE ranges do not overlap
SWE-bench 77.40% ±1.87 92.00% ±1.21 14.60 pts Reported ±1 SE ranges do not overlap
Terminal-Bench 4.0 11.11% ±1.34 37.37% ±2.67 26.26 pts Reported ±1 SE ranges do not overlap
Vibe Code Bench v1.1 49.93% ±4.77 30.76% ±5.28 19.17 pts Reported ±1 SE ranges do not overlap

Performance by category

Category DeepSeek V4 average GLM 5.3 Flash average
Legal 51.70% 64.56%
Finance 48.62% 53.57%
Healthcare 75.14% 88.94%
Math 16.00% 21.00%
Academic 88.32% 86.21%
Coding 46.20% 46.80%

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 cost GLM 5.3 Flash cost DeepSeek V4 latency GLM 5.3 Flash latency
Legal Research Bench $0.68 $0.24 17m05s 51m07s
LegalBench N/A N/A 3m04s 12.52s
Tax Agent Bench $0.72 $0.14 20m45s 21m37s
TaxEval v2 N/A N/A 3m10s 53.06s
MedScribe N/A N/A 5m46s 86.98s
ProofBench v1.1 $0.14 $0.45 7m03s 1h11m
GPQA Diamond N/A N/A 15m00s 2m11s
MMLU Pro N/A N/A 9m38s 44.76s
LiveCodeBench N/A N/A 11m56s 5m02s
ProgramBench N/A N/A 1h03m 5h30m
SkillsBench $0.40 $0.08 8m16s 12m09s
SWE-bench $0.44 $0.02 10m35s 1h04m
Terminal-Bench 4.0 $3.45 $1.22 1h26m 1h01m
Vibe Code Bench v1.1 $2.21 $4.70 56m59s 1h24m

Results available only for DeepSeek V4

  • MedCode
  • Public Benefits Bench v1.1

Results available only for GLM 5.3 Flash

  • Vals Index
  • Vals Finance Index
  • Vals Legal Index
  • HLAB-Vals-1.1
  • EMB v1.1
  • Finance Agent v2.1
  • Terminal-Bench-Science
  • MMMU Pro
  • SAGE
  • Code Migration v1.1
  • IOI
  • CyberBench v1.1
  • CyberOffense
  • SRE Bench
  • Access to Justice Benchmark
Model details DeepSeek V4 Model details GLM 5.3 Flash