Adds drift-diffusion response time families to 'frmtmb', for two-choice response time data. A noisy evidence accumulator runs between two boundaries and the response time is the first time it touches one; the drift rate, boundary separation, non-decision time and start point each take their own regression formula, and Ratcliff's across-trial variability in drift rate, start point and non-decision time is available as three more. The parameterization follows 'brms'. The Wiener family uses the Navarro and Fuss (2009) pair of series, written in plain R so that it differentiates exactly on an 'RTMB' tape, with a smooth weight between the two series because a tape cannot choose between them on a parameter. The generalized family of Shinn, Lam and Murray (2020) allows a state-dependent drift and moving boundaries, for which no closed-form density exists; it solves the Fokker-Planck equation on a grid that a change of variable holds fixed while the boundaries move, which is what keeps the likelihood differentiable. The linear ballistic accumulator of Brown and Heathcote (2008) is a race between any number of accumulators, and reaches choices with more than two alternatives, which a diffusion between two boundaries cannot. The racing diffusion model of Tillman, Van Zandt and Logan (2020) is the same race over noisy accumulators, so each finishing time is an inverse Gaussian averaged over the start point. The go/no-go model of Gomez, Ratcliff and Perea (2007) is the two-boundary diffusion observed at one boundary only, where a trial without a response carries the probability of no crossing before a deadline.