← Model indexMODEL PROFILE / deepseek-v4-pro

DeepSeek / current

DeepSeek V4 Pro

A long-context general model with thinking and non-thinking modes for reasoning, code and cost-efficient API work.

Context1M
ModalityText
AccessAPI
ReasoningHybrid

Where it fits.

Hybrid reasoning lets one model trade depth for latency by task, making it a flexible text-workload foundation. V4 Pro is DeepSeek's capability tier with 1.6T total and 49B active parameters. Official APIs provide 1M context, up to 384K output, thinking/non-thinking modes, JSON output and tool calls.

02

Core capabilities

  • 1M context and up to 384K output
  • Hybrid thinking and non-thinking operation
  • JSON output, tool calls and prefix completion
  • OpenAI Chat Completions and Anthropic-compatible interfaces
03

Best-fit workflows

  • Reasoning and mathematics
  • Code and tool calling
  • Long-context text workloads
04 / OPERATIONS

From evaluation
to production.

Deployment notes

Keep model IDs in configuration, not business logic. Use non-thinking for simple deterministic work and thinking for math, code or planning; cap output aggressively and validate tool arguments before execution.

Watch before choosing

Aliases and older names are retired over time; isolate model IDs from business logic. The old deepseek-chat and deepseek-reasoner aliases were scheduled for retirement in July 2026. Very long output and low token prices still require schema checks, truncation controls and workload-specific quality evaluation.

05

Family & versions

Specifications and positioning checked against provider documentation.

Open official source↗