The deregulated electricity market calls for robust OPF tools that can provide (a) deterministic convergence, (b) accurate computation of a variety of nodal prices, (c) support of continuous costs as well as discrete bids and offers, (d) full active and reactive power flow modeling of large-scale systems, and (e) satisfactory worst-case performance that meets the real-time dispatching requirement. For historical reasons, most prior research on OPF has focused on performance issues, without much treatment of requirements (a)-(c). This paper discusses these new challenges and presents two new algorithms for reliable computation of large-scale market-based OPF: trust-region based augmented Lagrangian method (TRALM) and step-controlled primal-dual interior point method (SCIPM). The former is more theoretically rigid while the latter is more effective in practice. The new algorithms, along with several existing ones, are tested and compared using largescale power system models.