Hill-type models are ubiquitous in the field of biomechanics, providing estimates of a muscle’s force as a function of its activation state and its assumed force-length and force-velocity properties. However, despite their routine use, the accuracy with which Hill-type models predict the forces generated by muscles during submaximal, dynamic tasks remains largely unknown. This study compared human gastrocnemii forces predicted by Hill-type models to the forces estimated from ultrasound-based measures of tendon length changes and stiffness during cycling, over a range of loads and cadences. We tested both a traditional model, with one contractile element, and a differential model, with two contractile elements that accounted for independent contributions of slow and fast muscle fibres. Both models were driven by subject-specific, ultrasound-based measures of fascicle lengths, velocities, and pennation angles and by activation patterns of slow and fast muscle fibres derived from surface electromyographic recordings. The models predicted on average, 54 % the time-varying gastrocnemii forces estimated from the ultrasound-based methods. However, differences between predicted and estimated forces were smaller under low speed-high activation conditions, with models able to predict nearly 80 % of the gastrocnemii force over a complete pedal cycle. Additionally, the predictions from the Hill-type muscle models tested here showed that a similar pattern of force production could be achieved for most conditions with and without accounting for the independent contributions of different muscle fibre types.