DeepSeek
DeepSeek V4 Pro (Reasoning, Max Effort)
DeepSeek V4 Pro (Reasoning, Max Effort) is a DeepSeek model profile, from a family known for strong reasoning, coding, and cost-conscious API models such as R1 and V-series releases. This page separates the public release context from measured benchmark performance so you can inspect its quality and efficiency directly.
DeepSeek release notesOperational Metrics
Output Speed34.3 tok/s
First Token1.24s
Blended Price$2.18/M
Model Metadata
Queryable facts extracted from the upstream model payload.
ReleaseApr 24, 2026
Context Windown/a
Modalitiesn/a
Strength: HLE
Rank #11 across 474 models.
35.9%
Strength: Overall
Rank #15 across 500 models.
51.5
Strength: GPQA
Rank #19 across 478 models.
88.8%
Watch Area: Speed
Rank #269 across 293 models.
34.3 tok/s
Watch Area: Input $
Rank #255 across 325 models.
$1.74/M
Watch Area: Blended $
Rank #227 across 325 models.
$2.18/M
Strength Profile
Percentile score by analysis domain.
* Cost is inverted: lower input, output, and blended prices rank higher.
Benchmark Percentiles
Higher bars mean stronger relative placement.
All Benchmarks
| Metric | Domain | Value | Rank |
|---|---|---|---|
| Artificial Analysis Intelligence Index | overall | 51.5 | #15 |
| Artificial Analysis Coding Index | coding | 47.5 | #17 |
| GPQA | reasoning | 88.8% | #19 |
| Humanity's Last Exam | reasoning | 35.9% | #11 |
| SciCode |