Home Model series DeepSeek V4 Pro DeepSeek-V4-Pro (Preview) DeepSeek-V4-Pro (Preview) · R&D Coding Benchmark (internal) · 30-task internal engineering evaluation; Pass Rate (%) · Scores and configurations DeepSeek-V4-Pro (Preview) · R&D Coding Benchmark (internal) · 30-task internal engineering evaluation; Pass Rate (%) · Scores and configurations DeepSeek-V4-Pro (Preview) record for R&D Coding Benchmark (internal) · 30-task internal engineering evaluation; Pass Rate (%). 1 score record for this release, including 1 exact value (may include historical records). Includes 1 source location.
Scores and evaluation conditions DeepSeek-V4-Pro (Preview) · R&D Coding Benchmark (internal) scores and configurations Metric and score Configuration and evaluation conditions Sources Pass Rate 67% Reported 67 · Unit % Published Unconfirmed
DeepSeek-V4-Pro (Preview) Max Settings Official mode label: Max; the row-specific inference budget and sampling parameters are not stated in this model-card table. Unknowns Do not assume local recommended parameters are the evaluation settings. Tools Tool access not recorded Reasoning budget Max Framework not recorded Protocol Official DeepSeek-V4 report Table 8; 30 internal R&D tasks with human-annotated rubrics. Conditions DeepSeek-V4-Pro-Max row; internal task set excluded from regular capability presentation. View this score in rankings
Benchmark versions and metrics Pass Rate · % · Higher is better