DeepSeek V4 Pro
deepseek/deepseek-v4-pro
Open MoE flagship with million-token context for coding and long agent runs
Capabilities
ReasoningYes
Tool callingYes
Structured outputYes
TemperatureSupported
AttachmentNo
Open weightsYes
Limits
Context1,000,000 tokens
Max inputN/A
Max output384,000 tokens
Details
Familydeepseek-thinking
Knowledge cutoff2025-05
Release date2026-04-24
Last updated2026-04-24
Modalities
Inputtext
Outputtext
Benchmarks
SWE-Bench Verified
80.6
resolved
Artificial Analysis Coding Agent Index
50.1
average pass@1
SWE-Atlas Codebase QnA
67.8
pass@1
SWE-Bench Pro
18
pass@1
Terminal-Bench
64.7
pass@1
Community Rating
—
No ratings yet
Sign in to rate
Pricing
Cost per 1M tokens
InputN/A
OutputN/A