[en] We compare simulations of the injection of mantle helium-3 into the deep ocean from six global coarse resolution models which participated in the Ocean Carbon Model Intercomparison Project (OCMIP). We also discuss the results of a study carried out with one of the models, which examines the effect of the subgrid-scale mixing parameterization. These sensitivity tests provide useful information to interpret the differences among the OCMIP models and between model simulations and the data. We find that the OCMIP models, which parameterize subgrid-scale mixing using an eddy-induced velocity, tend to underestimate the ventilation of the deep ocean, based on diagnostics with delta(3)He. In these models, this parameterization is implemented with a constant thickness diffusivity coefficient. In future simulations, we recommend using such a parameterization with spatially and temporally varying coefficients in order to moderate its effect on stratification. The performance of the models with regard to the formation of AABW confirms the conclusion from a previous evaluation with CFC-11. Models coupled with a sea-ice model produce a substantial bottom water formation in the Southern Ocean that tends to overestimate AABW ventilation, while models that are not coupled with a sea-ice model systematically underestimate the formation of AABW We also analyze specific features of the deep He-3 distribution (He-3 plumes) that are particularly well depicted in the data and which put severe constraints on the deep circulation. We show that all the models fail to reproduce a correct propagation of these plumes in the deep ocean. The resolution of the models may be too coarse to reproduce the strong and narrow currents in the deep ocean., and the models do not incorporate the geothermal heating that may also contribute to the generation of these currents. We also use the context of OCMIP-2 to explore the potential of mantle helium-3 as a tool to compare and evaluate modeled deep-ocean circulations. Although the source function of mantle helium is known with a rather large uncertainty, we find that the parameterization used for the injection of mantle helium-3 is sufficient to generate realistic results, even in the Atlantic Ocean where a previous pioneering study [J. Geophys. Res. 100 (1995) 3829] claimed this parameterization generates inadequate results. These results are supported by a multi-tracer evaluation performed by considering the simulated distributions of both helium-3 and natural C-14, and comparing the simulated tracer fields with available data. (C) 2004 Elsevier B.V. All rights reserved.