Head-to-Head: Clean Evals vs. Legacy Benchmarks
| Capability Domain | Clean Evals (Tiers S & A) | Legacy Benchmarks (Tiers C & D) | Public Lift (Boost) |
|---|---|---|---|
| Agentic & Enterprise | 46.3p | 17.0p | -29.3p |
| Coding & SWE | 45.1p | 24.5p | -20.6p |
| Math & Scientific Reasoning | 30.6p | 53.1p | +22.5p |
Non-comparable pre-release tests omitted from audited scoring (model released 2026-08-13): Terminal-Bench-Science 0.1 (2026-08-27).
Clean Evals Avg (Tiers S & A)
36.0p
10 private, leak-resistant tests
Legacy Benchmarks Avg (Tiers C & D)
31.5p
4 public or aging tests
Matched Public Lift (Boost)
-9.2p
Near zero or negative indicates clean generalization
Machine-Readable Export
Audited Benchmark Performance of DeepSeek V4 Pro 0813
Grouped by benchmark integrity so private capability signal stands apart from aging public tests.
Top-Tier Validated Tests (Tiers S & A) (10)
Private, leak-resistant tests that build real capability signal.
| Benchmark | Benchmark Integrity Status | Domain | Rank | Percentile |
|---|---|---|---|---|
| EnterpriseOps-Gym-AA | Private / Fresh | Agentic & Enterprise | #1/13 | 100.0p |
| AutomationBench-AA | Private / Fresh | Agentic & Enterprise | #36/50 | 28.6p |
| FrontierCode-AA 2.0 | Private / Fresh | Coding & SWE | #36/46 | 23.4p |
| FrontierMath | Private / Fresh | Math & Scientific Reasoning | #24/27 | 22.8p |
| AA-Briefcase v1.1 | Private / Fresh | Agentic & Enterprise | #45/50 | 10.2p |
| CyberBench v1.1 | Private / Fresh | Coding & SWE | #9/25 | 68.2p |
| Code Migration v1.1 | Private / Fresh | Coding & SWE | #14/24 | 43.8p |
| CritPt v2 | Private / Fresh | Math & Scientific Reasoning | #25/40 | 38.5p |
| ARC-AGI-2 | Private / Fresh | General Knowledge & QA | #18/22 | 28.2p |
| AA-Omniscience v1.1 | Private / Fresh | General Knowledge & QA | #40/47 | 15.3p |
Mid-Tier Evaluations (Tier B) (1)
Partially exposed or transitional evaluations with moderate leakage risk.
| Benchmark | Benchmark Integrity Status | Domain | Rank | Percentile |
|---|---|---|---|---|
| HLE-Diamond | Semi-Exposed | General Knowledge & QA | #35/49 | 29.2p |
Legacy & Vulnerable Benchmarks (Tiers C & D) (4)
Public or aging benchmarks prone to overfitting and training-data leaks.
| Benchmark | Benchmark Integrity Status | Domain | Rank | Percentile |
|---|---|---|---|---|
| GPQA Diamond | Public / Aging | Math & Scientific Reasoning | #8/17 | 53.1p |
| Terminal-Bench 4.0 | Public / Aging | Coding & SWE | #38/50 | 24.5p |
| GDPval-AA v2.1 | Public / Aging | Agentic & Enterprise | #33/43 | 23.8p |
| GDP.pdf | Public / Aging | Agentic & Enterprise | #45/50 | 10.2p |