deepseek-v4-pro

Common Name: DeepSeek V4 Pro

DeepSeek
Released on Aug 13 12:00 AMSupportedTool InvocationSupportedReasoning
CompareTry in Chat

DeepSeek V4 Pro is the higher-capability hybrid-thinking model in the V4 family. It supports both non-thinking and thinking modes, with 1M context and 384K max output.

Specifications

Context
1000K
Maximum Output
384K
Inputtext
Outputtext

Performance (7-day Average)

Collecting…
Collecting…
Collecting…

Pricing

Peak window (local): 01:00–04:00, 06:00–10:00

Off-peak
Input¥4.50/MTokens
Cached Input¥0.15/MTokens
Output¥13.50/MTokens
Peak
Input¥9.00/MTokens
Cached Input¥0.30/MTokens
Output¥27.00/MTokens

Performance Metrics (24h)

Similar Models

¥1.50/¥4.50/M
ctx1.0Mmax384Kavailtps
InOutCap

DeepSeek V4 Flash is a hybrid-thinking model with 1M context and 384K max output. It supports both non-thinking and thinking modes, with thinking enabled by default.

¥1.50/¥4.50/M
ctx1.0Mmax384Kavailtps
InOutCap

DeepSeek V4 Flash Vision Exp is the experimental vision variant of V4 Flash. Text capability matches the stable V4 Flash, and images are billed as token-equivalent. Supports 1M context and 384K max output.

$5.50/$33.00/M
ctx1.1Mmax128Kavailtps
InOutCap

GPT-5.6 is OpenAI's latest GPT-5.6 alias, routing requests to GPT-5.6 Sol for frontier capability across complex reasoning, coding, and production workflows.

$5.50/$33.00/M
ctx1.1Mmax128Kavailtps
InOutCap

GPT-5.6 Sol is the frontier model in the GPT-5.6 family for complex reasoning, coding, tool use, and demanding production workflows.