Unspecified · LIVEBENCH 2026-06-25

deepseek-v4-pro

Published benchmark results for this specific model configuration.

deepseek-v4-pro

Overall score71.6 / 100

Cost / successful task$0.050USD · benchmark workload

Weight accessNot reported

Where this configuration performs

Category averages from the same release, on a 0–100 scale.

Reasoning82.7

Coding70.0

Agentic Coding42.6

Mathematics90.7

Data Analysis74.5

Language78.1

Instruction Following62.4

Task-level results

theory of mind80.8zebra puzzle92.0spatial94.0logic with navigation64.0code generation70.4code completion69.6javascript54.5typescript23.3python50.0AMPS Hard98.0integrals with game78.0math comp96.1olympiad90.6consecutive events75.4tablejoin48.2tablereformat100.0connections98.0plot unscrambling56.4typos80.0paraphrase56.2simplify55.9story generation67.8summarize69.5

Source files: Scores ↗ Categories ↗ Costs ↗ . Missing cost is unknown, never free. Compare configurations within the same release.