Browser Compatibility

Core ML Backend / macOS

WebNN	CPU	GPU	NPU
macOS	✅ Core ML	✅ Core ML	✅ Core ML
iOS	❌ Core ML	❌ Core ML	❌ Core ML

The WebNN Core ML backend mainly supported on macOS.

WebNN Spec	Operations	Chromium Version
argMax	✅ reduce_argmax	M128
argMin	✅ reduce_argmin	M128
batchNormalization	✅ batch_norm	M126
cast	✅ cast	M124
clamp	✅ clip	M125
concat	✅ concat	M125
conv2d	✅ conv	M125
convTranspose2d	✅ conv_transpose	M126
cumulativeSum	✅ cumsum	M132
dequantizeLinear	🚀 dequantize_linear
element-wise binary / add	✅ add	M122
element-wise binary / div	✅ real_div	M124
element-wise binary / max	✅ maximum	M124
element-wise binary / min	✅ minimum	M124
element-wise binary / mul	✅ mul	M124
element-wise binary / notEqual	✅ not_equal	M134
element-wise binary / pow	✅ pow	M124
element-wise binary / sub	✅ sub	M124
element-wise logical / equal	✅ equal	M125
element-wise logical / greater	✅ greater	M125
element-wise logical / greaterOrEqual	✅ greater_equal	M125
element-wise logical / lesser	✅ less	M125
element-wise logical / lesserOrEqual	✅ less_equal	M125
element-wise logical / logicalAnd	✅ logical_and	M132
element-wise logical / logicalOr	✅ logical_or	M132
element-wise logical / logicalXor	✅ logical_xor	M132
element-wise logical / not	✅ logical_not	M128
element-wise unary / abs	✅ abs	M126
element-wise unary / ceil	✅ ceil	M126
element-wise unary / identity	✅ identity	M126
element-wise unary / cos	✅ cos	M126
element-wise unary / erf	✅ erf	M126
element-wise unary / exp	✅ exp	M126
element-wise unary / floor	✅ floor	M126
element-wise unary / log	✅ log	M126
element-wise unary / neg	✅ Emulated with mul(a, -1)	M127
element-wise unary / reciprocal	✅ inverse	M126
element-wise unary / sin	✅ sin	M126
element-wise unary / sqrt	✅ sqrt	M126
element-wise unary / tan	✅ tan	M126
elu	✅ elu	M126
expand	✅ tile	M128
gather	✅ gather	M126
gatherElements	✅ gather_along_axis	M132
gatherND	✅ gather_nd	M132
gelu	✅ gelu	M132
gemm	✅ Emulated with add(mul(alpha, matmul(A, B)), mul(beta, C))	M126
gru	✅ gru	M133
gruCell	✅ gru	M133
hardSigmoid	✅ sigmoid_hard	M126
hardSwish	✅ Emulated by mul(x, hardsigmoid(x, alpha=1.0/6, beta=0.5))	M126
instanceNormalization	✅ instance_norm	M126
layerNormalization	✅ layer_norm	M129
leakyRelu	✅ leaky_relu	M126
linear	Emulated with add(mul(alpha, a), beta)	M126
lstm	✅ lstm	M133
lstmCell	✅ lstm	M133
matmul	✅ matmul	M126
pad	✅ pad	M129
pooling / averagePool2d	✅ avg_pool	M126
pooling / l2Pool2d	✅ l2_pool	M126
pooling / maxPool2d	✅ max_pool	M126
prelu	🚀 prelu
quantizeLinear	🚀 quantize_linear
reduction / reduceL1	✅ reduce_l1_norm	M126
reduction / reduceL2	✅ reduce_l2_norm	M126
reduction / reduceLogSum	✅ reduce_log_sum	M126
reduction / reduceLogSumExp	✅ reduce_log_sum_exp	M126
reduction / reduceMax	✅ reduce_max	M126
reduction / reduceMean	✅ reduce_mean	M126
reduction / reduceMin	✅ reduce_min	M126
reduction / reduceProduct	✅ reduce_prod	M126
reduction / reduceSum	✅ reduce_sum	M126
reduction / reduceSumSquare	✅ reduce_sum_square	M126
relu	✅ relu	M125
resample2d	✅ upsample_bilinear ✅ upsample_nearest_neighbor	M126
reshape	✅ reshape	M126
reverse	🚀 reverse
scatterElements	✅ scatter_along_axis	M132
scatterND	✅ scatter_nd	M132
sigmoid	✅ sigmoid	M126
sign	🚀 sign
slice	✅ slice_by_size	M126
softmax	✅ softmax	M127
softplus	✅ softplus	M126
softsign	✅ softsign	M125
split	✅ split	M130
tanh	✅ tanh	M126
tile	✅ tile	M131
transpose	✅ transpose	M125
triangular	✅ band_part	M132
where	✅ select	M126

DirectML Backend WebNN EP Ops