This article presents a versatile single-photon-based quantum computing platform called Ascella, which is cloud-accessible and capable of performing both gate-based and photonic computation tasks. The platform uses six photonic qubits generated by an on-demand quantum dot source, with quantum information encoded in the path degree of freedom. A 12-mode integrated universal interferometer is used to implement all 12 × 12 unitary matrices with high fidelity. A machine-learned transpilation process corrects for hardware manufacturing errors, enabling high-performance operation with a six-photon sampling rate of 4 Hz and stability over weeks. The platform demonstrates state-of-the-art performance in gate-based quantum computation, achieving fidelities of 99.6 ± 0.1%, 93.8 ± 0.6%, and 86 ± 1.2% for one-, two-, and three-qubit gates, respectively. It also implements a variational quantum eigensolver (VQE) to calculate the energy levels of the hydrogen molecule with chemical accuracy. For photonic computation, the platform implements a quantum neural network classifier using a three-photon-based quantum neural network and reports a six-photon boson sampling demonstration on a universal reconfigurable integrated circuit. The platform also demonstrates heralded three-photon entanglement generation, a key milestone toward measurement-based quantum computing. The results show that the platform can perform complex quantum computations with high fidelity and stability, and it has the potential for future scaling to larger systems. The platform's architecture and performance are compared to other quantum computing platforms, and it is shown to be competitive with superconducting and ion trap quantum computing systems. The study highlights the potential of single-photon-based quantum computing for near-term quantum computing tasks and for large-scale fault-tolerant quantum computing.This article presents a versatile single-photon-based quantum computing platform called Ascella, which is cloud-accessible and capable of performing both gate-based and photonic computation tasks. The platform uses six photonic qubits generated by an on-demand quantum dot source, with quantum information encoded in the path degree of freedom. A 12-mode integrated universal interferometer is used to implement all 12 × 12 unitary matrices with high fidelity. A machine-learned transpilation process corrects for hardware manufacturing errors, enabling high-performance operation with a six-photon sampling rate of 4 Hz and stability over weeks. The platform demonstrates state-of-the-art performance in gate-based quantum computation, achieving fidelities of 99.6 ± 0.1%, 93.8 ± 0.6%, and 86 ± 1.2% for one-, two-, and three-qubit gates, respectively. It also implements a variational quantum eigensolver (VQE) to calculate the energy levels of the hydrogen molecule with chemical accuracy. For photonic computation, the platform implements a quantum neural network classifier using a three-photon-based quantum neural network and reports a six-photon boson sampling demonstration on a universal reconfigurable integrated circuit. The platform also demonstrates heralded three-photon entanglement generation, a key milestone toward measurement-based quantum computing. The results show that the platform can perform complex quantum computations with high fidelity and stability, and it has the potential for future scaling to larger systems. The platform's architecture and performance are compared to other quantum computing platforms, and it is shown to be competitive with superconducting and ion trap quantum computing systems. The study highlights the potential of single-photon-based quantum computing for near-term quantum computing tasks and for large-scale fault-tolerant quantum computing.