DeepSeek V4 Pro
DeepSeek · released
The general-availability release of DeepSeek's larger V4 model: text-only, 1.6 trillion parameters, tuned for agent and coding work. It is still sold, although DeepSeek says the newer V4.1 Flash beats it and a V4.1 Pro is planned.
Best for: MIT-licensed open weights · Text-only agents · Self-hosting at scale
- Intelligence Index
- 36.0
- #23 of 38 · at Max effort
- Price per 1M tokens
- $1.32 in $3.96 out
- Blended $1.98 at 3:1
- Context window
- 1M
- Up to 384K output
- Output speed
- 100 tokens/s
- $0.67 measured cost per task
Capability profile
One 0–100 score per area, placing this model between the weakest and strongest tracked model across that area's benchmarks.
- Overall56 · #22 of 38
Based on 3 of 3 benchmarks
- Coding49 · #23 of 39
Based on 5 of 5 benchmarks
- Agents & knowledge work55 · #25 of 40
Based on 6 of 6 benchmarks
- Reasoning & science59 · #18 of 39
Based on 2 of 2 benchmarks
- Knowledge & reliability63 · #21 of 39
Based on 2 of 2 benchmarks
- VisionNot enough data
Based on 0 of 2 benchmarks
Reasoning effort trade-off
The same model at each effort setting. Lower settings give up a few points for a fraction of the cost and wait.
| Setting | Intelligence Index | Cost / task | First answer |
|---|---|---|---|
| Max | 36.0 | $0.67 | 22s |
| No reasoning | 20.4 | $0.46 | 1.6s |
Measured by Artificial Analysis. First answer is the median wait until the first answer token, including thinking time.
Benchmark results
All third-party results on record. Rank is among the tracked models that have been tested on that benchmark.
Overall
#23 of 38· Max effort· Artificial Analysis
- Vals Index47.6%
#22 of 26· Max effort· $3.35 per test· Vals AI
- Arena · Text1464 ±7
#17 of 33· 9,780 votes· Arena
Coding
- Terminal-Bench 4.014.1%
#21 of 38· Max effort· Artificial Analysis
- SciCode51.0%
#23 of 38· Max effort· Artificial Analysis
- Arena · WebDev1582 ±9
#20 of 34· 4,744 votes· Arena
- Vibe Code Bench82.3%
#13 of 26· Max effort· Vals AI
- Code Migration41.6%
#16 of 26· Max effort· Vals AI
Agents & knowledge work
- GDPval-AA47.7
#19 of 39· Max effort· Artificial Analysis
- Arena · Agent+1.2 ±1.7
#17 of 29· High effort· Arena
- Finance Agent v250.4%
#21 of 26· Max effort· Vals AI
#22 of 26· Max effort· Vals AI
- Legal Research Bench40.9%
#15 of 26· Max effort· Vals AI
- Tax Agent Bench58.7%
#22 of 26· Max effort· Vals AI
Reasoning & science
- Humanity's Last Exam41.0%
#18 of 39· Max effort· Artificial Analysis
- CritPt18.0%
#16 of 39· Max effort· Artificial Analysis
- ARC-AGI-2saturating61.3%
#11 of 16· Max effort· $0.60 per task· ARC Prize
- ARC-AGI-3Not tested
- GPQA Diamondsaturating92.8%
#8 of 27· Max effort· Artificial Analysis
Knowledge & reliability
#20 of 39· Max effort· Artificial Analysis
- AA-LCR80.3%
#20 of 39· Max effort· Artificial Analysis
Vision
- MMMU-ProNot tested
- Arena · VisionNot tested
Reported by DeepSeek
Numbers the lab published at launch, run on its own harness and settings. Useful context, but not independently verified and never used in the rankings on this site.
- DeepSWE v1.1 · mini-SWE harness62.7%
- AutomationBench43.2%
- GPQA Diamond92.4%
Specifications
- Lab
- DeepSeek
- Released
- Aug 13, 2026
- Availability
- Generally available
- Context window
- 1,000,000 tokens
- Max output
- 384,000 tokens
- Accepts
- Text
- Knowledge cutoff
- Not published
- Weights
- Open
- Licence
- MIT
- Parameters
- 1.6T total · 49B active
- Reasoning control
- Thinking on by default · can be switched off
- API model ID
deepseek-v4-pro
Pricing
- Input / 1M tokens
- $1.32
- Cached input / 1M
- $0.04
- Output / 1M tokens
- $3.96
- Blended (3:1)
- $1.98
- Measured cost / task
- $0.67
Peak-hour list price. Off-peak requests cost half.
Official sources
Closest alternatives
The models nearest to this one on overall score.
23 benchmarks tracked. See methodology and sources.