Orthogonal frequency division multiplexing with index modulated (OFDM-IM) is a novel energy efficient variant of OFDM that conveys information through active subcarrier indices and conventional M-ary symbols. OFDM-IM’s potential to be spectrally efficient has been explored extensively in prior work. However, the spectral efficiency (SE) metric is only achievable for very high SNR. For general channel conditions, the achievable rates determined using information-theoretic limits are likely lower than the SE. In this paper, we propose a mapping scheme for an OFDM-IM system that optimizes the mutual information. The proposed scheme uses OFDM-IM symbols with different probabilities of occurring, based on a dyadic probability distribution that best approximates the capacity-achieving input distribution for a peak-power-constrained AWGN channel at a given SNR. The dyadic probability distribution is based on Huffman coding, which is used to create the mapping scheme from OFDM-IM constellation symbols to variable-length bit sequences. The simulation results show that for the same number of subchannels and constellation size, the proposed scheme achieves better mutual information than existing OFDM-based schemes for higher modulation orders.