A current bottleneck for quantum computation is the realization of high-fidelity two-qubit quantum operations between two and more quantum bits in arrays of coupled qubits. Gates based on parametrically driven tunable couplers offer a convenient method to entangle multiple qubits by selectively activating different interaction terms in the effective Hamiltonian. Here, we study theoretically and experimentally a superconducting qubit setup with two transmon qubits connected via a capacitively coupled tunable bus. We develop a time-dependent Schrieffer-Wolff transformation and derive analytic expressions for exchange-interaction gates swapping excitations between the qubits (iSWAP) and for two-photon gates creating and annihilating simultaneous two-qubit excitations (bSWAP). We find that the bSWAP gate is generally slower than the more commonly used iSWAP gate, but features favorable scalability properties with less severe frequency crowding effects, which typically degrade the fidelity in multi-qubit setups. Our theoretical results are backed by experimental measurements as well as exact numerical simulations including the effects of higher transmon levels and dissipation.Quantum computation is based on accurate and precise control of quantum bits and their interactions to create multi-qubit superpositions and entanglement. With superconducting circuits, single qubit quantum gates can be carried out with fidelities approaching 99.99% [1-4], while errors in two-qubit operations are typically higher with record fidelities around 99% [3,5]. However, the realization of qubit operations with even higher fidelity is required both for reaching the error threshold for quantum computation [6-9] and for carrying out reliable quantum simulations and optimizations in large arrays of coupled qubits [10][11][12][13]. Moreover, the quest for useful quantum computations before full quantum error correction becomes available may be assisted by efficient, short-depth gate sequences based on two-or multi-qubit gates [14,15] with versatile types of interactions. In particular, parametric schemes based on tunable couplers have been proposed and recently realized as a means to achieve fast gates with high fidelities [16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31].In this context, effective interactions were engineered in Ref.[26] between two transmon qubits mediated by a third, ancilla transmon device (bus), which couples dispersively to both qubits and whose frequency is modulated by an external magnetic flux. Such a flux-modulation scheme provides frequency-selectivity and allows to use fixed-frequency computational qubits, thereby minimizing the sensitvity of the device with respect to magnetic flux noise and disorder effects. For example, modulating at the (fixed) difference frequency of the qubits brings these qubits effectively into resonance in a co-rotating frame such that a single excitation can be swapped efficiently (iSWAP). The effective Hamiltonian in this case is H iSWAP ∝ XX + YY. With this method gate ...