Code Generation on HumanEval, HumanEval+, MBPP, MBPP+, Bigcode (test) benchmark leaderboard · SOTA2 Research

Code Generation on HumanEval, HumanEval+, MBPP, MBPP+, Bigcode (test)

48.2HumanEvalSelectIT

Evaluation Results

Method	Links
SelectIT2026.01		48.2	38.4	60.6	51.9	23	44.4
InstructDiff2026.01		48.2	40.9	60.6	50	25.7	45.1
Resp LenMax2026.01		47	37.8	57.4	48.4	24.2	43
Inst/RespMin2026.01		47	38.4	56.1	46.3	22.8	42.1
EntropyMin2026.01		46.3	39	58.7	46.3	23.4	42.7
ALL2026.01		44.5	35.4	59.8	49.7	25.8	43
PPLMid2026.01		44.5	36.6	56.9	46.8	21.6	41.3
Inst/RespMax2026.01		43.9	38.4	59	49.2	21.9	42.5
Superfiltering2026.01		43.3	34.8	54.2	44.2	20.6	39.4
Inst LenMax2026.01		42.1	35.4	57.7	46.3	23.6	41
Random2026.01		42.1	33.5	57.7	47.9	22.9	40.8
PPLMin2026.01		41.5	34.1	56.1	46.3	23.4	40.3
EntropyMid2026.01		41.5	34.1	59.3	48.1	23.8	41.4
EntropyMax2026.01		40.9	34.8	57.9	48.4	21.1	40.6
IFD2026.01		40.9	34.1	51.6	40.7	18.2	37.1
Base Model2026.01		40.2	33.5	55.3	46.3	22.5	39.6
PPLMax2026.01		40.2	34.1	55.6	47.4	21	39.7
ZIP2026.01		38.4	32.9	59.3	50.3	16	39.4