Abstract. In this paper, we derive a general asymptotic implied volatility at the first-order for any stochastic volatility model using the heat kernel expansion on a Riemann manifold endowed with an Abelian connection. This formula is particularly useful for the calibration procedure. As an application, we obtain an asymptotic smile for a SABR model with a mean-reversion term, called λ-SABR, corresponding in our geometric framework to the Poincaré hyperbolic plane. When the λ-SABR model degenerates into the SABR-model, we show that our asymptotic implied volatility is a better approximation than the classical Haganal expression [19]. Furthermore, in order to show the strength of this geometric framework, we give an exact solution of the SABR model with β = 0 or 1. In a next paper, we will show how our method can be applied in other contexts such as the derivation of an asymptotic implied volatility for a Libor market model with a stochastic volatility ([23]).