Home Model series DeepSeek R1 Zero DeepSeek-R1-Zero DeepSeek-R1-Zero · MMLU-Pro · DeepSeek-R1 paper Table 3; MMLU-Pro (EM); unit not printed · Scores and configurations DeepSeek-R1-Zero · MMLU-Pro · DeepSeek-R1 paper Table 3; MMLU-Pro (EM); unit not printed · Scores and configurations DeepSeek-R1-Zero record for MMLU-Pro · DeepSeek-R1 paper Table 3; MMLU-Pro (EM); unit not printed. 1 score record for this release, including 1 exact value (may include historical records). Includes 1 source location.
Scores and evaluation conditions DeepSeek-R1-Zero · MMLU-Pro scores and configurations Metric and score Configuration and evaluation conditions Sources Exact Match 68.9 Reported 68.9 · Unit not specified Published Unconfirmed
DeepSeek-R1-Zero reported evaluation Settings Pure-RL R1-Zero checkpoint; report protocol uses sampled Pass@1 with maximum generation length 32,768 and row-specific benchmark settings. Unknowns The paper table does not print percent signs.、Per-row response counts are not repeated in Table 3. Tools Tool access not recorded Reasoning budget Output limit 32768 tokens Framework not recorded Protocol Official DeepSeek-R1 paper Table 3; R1-Zero stage evaluation with sampled reasoning configuration. Conditions R1-Zero column; complete row retained. The table does not print percent signs. View this score in rankings
Benchmark versions and metrics Exact Match · Unit not specified · Higher is better