Kimi-K2-Base at launch
Independent boards have not scored it on enough tests for the LLMBenchmark Index yet; below are the results Moonshot AI reported and the independent scores published so far. Moonshot AI's own announcement reports 13 benchmark results, including GPQA Diamond (48.1%) and MMLU Pro (69.2%).
What Moonshot AI reported at launch
From Moonshot AI's announcement. The maker's own numbers, not re-run by us.
-
C-Eval
92.5
Base model evaluation results; EM
-
CSimpleQA
77.6
Base model evaluation results; Correct
-
GSM8K
92.1
Base model evaluation results; EM
#2 of 2 models with a reported result
-
MMLU
87.8%
Base model evaluation results; EM
#3 of 9 models with a reported result
-
EvalPlus
80.3
Base model evaluation results; Pass@1
-
GPQA Diamond
48.1%
Base model evaluation results; Avg@8
#49 of 50 models with a reported result
-
LiveCodeBench v6
26.3
Base model evaluation results; Pass@1
#14 of 14 models with a reported result
-
MATH
70.2
Base model evaluation results; EM
#1 of 4 models with a reported result
-
MMLU Pro
69.2%
Base model evaluation results; EM
#16 of 20 models with a reported result
-
MMLU-redux-2.0
90.2%
Base model evaluation results; EM
-
SimpleQA
35.3
Base model evaluation results; Correct
#3 of 7 models with a reported result
-
SuperGPQA
44.7
Base model evaluation results; EM
#4 of 4 models with a reported result
Show all 13
-
TriviaQA
85.1
Base model evaluation results; EM
Independent results so far
Not published