DeepSeek V4.1 Flash vs GPT-5.6 Terra: Benchmark Comparison

DeepSeek V4.1 Flash has the higher score on 6 of 17 shared benchmarks; GPT-5.6 Terra leads on 9.

The largest observed score gap is 47.33 pts on IOI , where GPT-5.6 Terra leads.

Reported ±1 standard-error ranges overlap on 10 of 17 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 V4.1 Flash GPT-5.6 Terra Gap Reported uncertainty
Legal Research Bench 41.35% ±3.42 41.35% ±3.42 0.00 pts Reported ±1 SE ranges overlap
LegalBench 83.28% ±0.46 85.11% ±0.45 1.83 pts Reported ±1 SE ranges do not overlap
Tax Agent Bench 26.63% ±3.14 30.04% ±3.31 3.41 pts Reported ±1 SE ranges overlap
MedCode 41.17% ±2.04 43.41% ±2.17 2.24 pts Reported ±1 SE ranges overlap
MedScribe 85.50% ±1.92 82.87% ±1.95 2.63 pts Reported ±1 SE ranges overlap
ProofBench v1.1 54.00% ±5.01 74.00% ±4.41 20.00 pts Reported ±1 SE ranges do not overlap
Terminal-Bench-Science 0.00% ±0.00 10.00% ±3.61 10.00 pts Reported ±1 SE ranges do not overlap
SAGE 85.38% ±1.96 87.04% ±2.00 1.65 pts Reported ±1 SE ranges overlap
IOI 40.28% ±2.56 87.61% ±6.53 47.33 pts Reported ±1 SE ranges do not overlap
ProgramBench 0.50% ±0.50 0.50% ±0.50 0.00 pts Reported ±1 SE ranges overlap
SkillsBench 69.80% ±3.88 58.90% ±4.47 10.91 pts Reported ±1 SE ranges do not overlap
Terminal-Bench 4.0 19.70% ±1.75 22.73% ±2.31 3.03 pts Reported ±1 SE ranges overlap
Vibe Code Bench v1.1 84.74% ±2.89 74.59% ±4.20 10.15 pts Reported ±1 SE ranges do not overlap
CyberBench v1.1 73.69% ±5.48 72.08% ±5.41 1.61 pts Reported ±1 SE ranges overlap
SRE Bench 0.76% ±0.54 9.54% ±1.82 8.78 pts Reported ±1 SE ranges do not overlap
Financial Coaching Bench 72.83% ±1.57 70.98% ±1.54 1.86 pts Reported ±1 SE ranges overlap
Public Benefits Bench v1.1 64.28% ±1.25 62.38% ±1.26 1.89 pts Reported ±1 SE ranges overlap

Performance by category

Category DeepSeek V4.1 Flash average GPT-5.6 Terra average
Legal 62.31% 63.23%
Finance 26.63% 30.04%
Healthcare 63.34% 63.14%
Math 54.00% 74.00%
Science 0.00% 10.00%
Education 85.38% 87.04%
Coding 43.00% 48.87%
Cyber 37.23% 40.81%
Social Mobility 68.55% 66.68%

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 V4.1 Flash cost GPT-5.6 Terra cost DeepSeek V4.1 Flash latency GPT-5.6 Terra latency
Legal Research Bench $0.25 $7.91 11m42s 1h10m
LegalBench N/A N/A 6.53s 2.78s
Tax Agent Bench $0.13 $4.83 5m49s 1h01m
MedCode N/A N/A 35.13s 18.41s
MedScribe N/A N/A 53.83s 35.53s
ProofBench v1.1 $0.13 $1.27 9m07s 12m57s
Terminal-Bench-Science $0.47 $5.19 2h57m 2h04m
SAGE $0.01 $0.11 33.11s 42.92s
IOI $0.24 $8.66 18m30s 1h23m
ProgramBench $0.90 $5.38 1h02m 33m36s
SkillsBench $0.09 $1.75 3m43s 6m55s
Terminal-Bench 4.0 $0.50 $5.60 48m22s 31m29s
Vibe Code Bench v1.1 $0.41 $7.89 15m28s 19m12s
CyberBench v1.1 $0.11 $3.31 13m50s 17m04s
SRE Bench $0.55 $30.86 1h01m 3h17m
Financial Coaching Bench $0.02 $0.31 6m08s 5m22s
Public Benefits Bench v1.1 $0.07 $1.20 9m59s 19m31s

Results available only for DeepSeek V4.1 Flash

  • Vals Finance Index
  • Vals Science Index
  • Vals Inference Provider Index
  • Immigration
  • EMB v1.1
  • Finance Agent v2.1
  • BioMysteryBench
  • MysteryMechanism
  • Code Migration v1.1
  • SWE-bench Multimodal
  • Access to Justice Benchmark

Results available only for GPT-5.6 Terra

  • MortgageTax
  • TaxEval v2
  • GPQA Diamond
  • MMLU Pro
  • MMMU Pro
  • LiveCodeBench
  • SWE-bench
Model details DeepSeek V4.1 Flash Model details GPT-5.6 Terra