Private AI
DeepSeek V4 Pro 0813 Thinking enables reasoning by default on the general-availability release of DeepSeek V4 Pro, built for coding, tool use, cybersecurity, automation, and long-horizon agent workflows. DeepSeek reports strong gains over the preview and scores above Opus 4.8 on Terminal Bench 2.1, Cybergym, DeepSWE, and AutomationBench. It supports a 1M-token context window. For now, this model routes directly through DeepSeek, which may log and train on your prompts.
Added Aug 12, 2026
Context Window
1.0M
Max Output
384.0K
Input Price (Auto)
$0.43/1M
Output Price (Auto)
$0.87/1M
Cache Read (Auto)
$0.0036/1M
Capabilities
Performance metrics and benchmarks
Sourced from Artificial Analysis.
Intelligence Index
45.3
Coding Index
59.4
Agentic Index
37.8
GPQA Diamond
Graduate-level scientific reasoning
88.8%
Better than 89% of models compared
HLE
Humanity's Last Exam
37.5%
Last updated Aug 12, 2026
Artificial AnalysisAuto routing is available for this model. Explicit provider selection is not available.
Loading provider options…
Better than 92% of models compared
IFBench
Instruction-following benchmark
76.5%
Better than 95% of models compared
T²-Bench Telecom
Conversational AI agents in dual-control scenarios
96.2%
Better than 97% of models compared
AA-LCR
Long context reasoning evaluation
70.0%
Better than 80% of models compared
GDPval-AA
Economically valuable tasks
40.3%
CritPt
Research-level physics reasoning
12.9%
SciCode
Python programming for scientific computing
50.0%
Better than 91% of models compared
Terminal-Bench Hard
Agentic coding and terminal use
46.2%
Better than 93% of models compared
AA-Omniscience Accuracy
Proportion of correctly answered questions
42.9%
AA-Omniscience Hallucination Rate
Rate of incorrect answers among non-correct responses
94.1%
