deepseek-v4-pro:cloud

Cloud tags

Key features

• Frontier reasoning — 1.6T total parameters with 49B activated, tuned for frontier-level reasoning.
• Million-token context — a 1M-token context window across the full DeepSeek-V4 series.
• Three thinking modes — No thinking for fast, intuitive answers; Thinking for careful logical analysis; Max thinking for maximum reasoning effort on the hardest problems.

Best practices

• Thinking modes — pick No thinking for quick answers, Thinking for logical analysis, and Max thinking only for the hardest problems to balance quality and latency.
• Long context — the 1M window suits corpus-scale analysis and long agentic trajectories.
• Flagship tier — the extra-high usage level reflects peak demand; prefer V4-Flash for cost-efficient workloads.