Overall score
50.6
Rank #57 · A confidence
DeepSeek · preview
DeepSeek's official lower-cost V4 API model with optional thinking and open model weights.
Overall score
50.6
Rank #57 · A confidence
Capability
39.6
21 eligible benchmark families
Coverage
81%
Confidence A
Benchmarks scored
21
25 raw result rows
Model specification
Identity, configuration and price are preserved as dated source records.
Category profile
A category remains Not evaluated until an eligible source-qualified result is attached.
Benchmark scores
Best published score per benchmark. Open any row for the full per-benchmark leaderboard.
| Benchmark | Category | Score | Date |
|---|---|---|---|
| τ²-Bench TelecomRanking weighted | agents | 95.0292% | 2026-07-15 |
| GPQA DiamondRanking weighted | reasoning | 89.39% | 2026-07-15 |
| MMLU-ProRanking weighted | research | 83% | 2026-07-15 |
| SWE-bench VerifiedRanking weighted | coding | 73.7% | 2026-07-15 |
| MCP AtlasRanking weighted | agents | 64% | 2026-07-15 |
| AA Long Context ReasoningRanking weighted | research | 63% | 2026-07-15 |
| Terminal-BenchRanking weighted | agents | 61.8% | 2026-07-15 |
| LiveCodeBenchRanking weighted | coding | 55.2% | 2026-07-15 |
| SWE-Bench ProRanking weighted | coding | 49.1% | 2026-07-15 |
| HMMT Feb 2026Ranking weighted | mathematics | 40.8% | 2026-07-15 |
| ToolathlonRanking weighted | agents | 40.7% | 2026-07-15 |
| EnterpriseOps-Gym-AARanking weighted | agents | 39.5703% | 2026-07-15 |
| Terminal-Bench HardRanking weighted | agents | 35.6061% | 2026-07-15 |
| GDPval-AA v2Ranking weighted | agents | 34.427% | 2026-07-15 |
| Humanity's Last ExamRanking weighted | reasoning | 32.07% | 2026-07-15 |
| ITBench-AARanking weighted | agents | 31.516% | 2026-07-15 |
| SimpleQARanking weighted | research | 23.1% | 2026-07-15 |
| τ³-BankingRanking weighted | agents | 22.8866% | 2026-07-15 |
| CritPtRanking weighted | reasoning | 7.1429% | 2026-07-15 |
| Harvey LAB-AARanking weighted | agents | 1.6667% | 2026-07-15 |
| AA-OmniscienceRanking weighted | research | -22.9 index | 2026-07-15 |
Showing the best of 25 raw result rows across 21 benchmark families.
Price and efficiency
Operational values are shown only when the source publishes a comparable measurement or a valid derivation.
Verification summary
News and X posts are structurally excluded from this model's evidence graph.
Missing publication dimensions: value. The last source check for this exact model was 2026-07-14.
Related models
Exact variants remain separate even when they share a provider or family.
DeepSeek V4 · preview
DeepSeek V3.2DeepSeek V3 · stable
DeepSeek V3.1 TerminusDeepSeek V3 · stable