General theoretical benefits:

• captures neural representation of arbitrarily complex objects (i.e. scalars, vectors, functions, vector fields, etc.)
• captures linear and nonlinear neural computations
• captures arbitrary neural dynamics (linear, nonlinear, time-varying, adaptive, etc.)
• unifies population and temporal coding
• quantitatively specified

Neural modelling benefits:

• analytically determines connection weights for models (this can be reversed; i.e. given weights, high-level transformations can be extracted)
• incorporates any available neural data (e.g., tuning curves, response functions, trasmitter/receptor types, anatomical constraints, etc.)
• appropriate regardless of single cell model (i.e. rate, spiking, full conductance based)
• arbitrary network topologies
• complexity of models is smoothly variable (i.e. single cells, population level, systems level)
• learning is not need to construct informative models, but can be included as desired
• means to test abstract hypotheses about neural function
• relates directly to recorded neural data (e.g. spikes, tuning curves, pooled spikes, fMRI signals (i.e. filtered spiking activity), LFP, EEG, etc.)

Simulation benefits:

• easily scales complexity to allow match to available hardware
• no learning is necessary (though allowed), permitting rapid model development
• GUI simulator freely available, with full scripting (Nengo)