Exciting progress towards spin-based quantum computing [1,2] has recently been made with qubits realized using nitrogen-vacancy (N-V) centers in diamond and phosphorus atoms in silicon [3], including the demonstration of long coherence times made possible by the presence of spin-free isotopes of carbon [4] and silicon [5]. However, despite promising single-atom nanotechnologies [6], there remain substantial challenges in coupling such qubits and addressing them individually. Conversely, lithographically defined quantum dots have an exchange coupling that can be precisely engineered1, but strong coupling to noise has severely limited their dephasing times and control fidelities. Here we combine the best aspects of both spin qubit schemes and demonstrate a gate-addressable quantum dot qubit in isotopically engineered silicon with a control fidelity of 99.6%, obtained via Clifford based randomized benchmarking and consistent with that required for fault-tolerant quantum computing [7,8]. This qubit has orders of magnitude improved coherence times compared with other quantum dot qubits, with T * 2 = 120 µs and T2 = 28 ms. By gate-voltage tuning of the electron g * -factor, we can Stark shift the electron spin resonance (ESR) frequency by more than 3000 times the 2.4 kHz ESR linewidth, providing a direct path to large-scale arrays of addressable high-fidelity qubits that are compatible with existing manufacturing technologies.The seminal work by Loss and DiVincenzo[1] to encode quantum information using the spin states of semiconductor quantum dots generated great excitement, as it fulfilled what were then understood to be the key criteria[2] for quantum computation, and has already led to the realization of 2-qubit operations such as the √ SWAP[9, 10] and CPHASE [11]. However, the limited lifetime and the associated fidelity of the quantum state represent a significant hurdle for the semiconductor quantum dot qubits realized thus far. A dephasing time up to T * 2 = 37 ns [12], improved to T * 2 = 94 ns[13] using nuclear spin bath control, has been recorded for quantum dot spin qubits in GaAs/AlGaAs. A longer T * 2 = 360 ns has been achieved using Si/SiGe quantum dots [14]. The main strategy to improve these times has involved applying pulse sequences developed for bulk magnetic resonance, and we can specify a T 2 according to the applied pulse sequence. Using a Hahn echo sequence the coherence time of GaAs-based qubits has been extended to T 28 Si), we remove the dephasing effect of the nuclear spin bath present in these previous studies, and show that all of the above coherence times can be improved by orders of magnitude. These long coherence times, in particular the dephasing time T * 2 , lead to low control error rates and the high fidelities that will be required for large-scale, fault tolerant quantum computing [7,8].In contrast with quantum dots, electron spin qubits localized on atoms or defects have been realized in almost spinfree environments, showing coherence times approaching [4] and even exceeding seco...