According to motor control models based on optimal planning, the generation of a sequence of skilled movements is based on the selection of optimal motor commands by the Central Nervous System. This concept has been implemented in a first version of our speech production model GEPPETO by minimizing the size of the neighborhood that includes all the command values specifying the successive phoneme-related targets in a speech sequence. We propose an improvement of this model in order to be able to deal with time and gesture accuracy constraints. The optimal planning is now achieved under a double constraint, namely the achievement of the perceptual target (as in the original model) and the limitation of the global muscle force level within a given range during the whole tongue movement. The selected force range is guided by speaking rate and by perceptual accuracy requirements: for slow speaking rates or low accuracy requirements, a low level of force can be used; for fast speaking rates and great accuracy requirement a strong level of force is required. The new optimal planning process has been tested on short speech sequences. Optimal motor command patterns were found for each segment of the sequences for each of the three force constraints (low, medium and strong force). Then, a biomechanical tongue model was used to simulate movements for every force level and for three timing of the motor commands: slow, normal and fast. The results show that the global muscle force can have a significant impact on the articulatory trajectories, in terms of curvature and in terms of positions at targets.