The NASA Carbon Monitoring System (CMS) Flux Project aims to attribute changes in the atmospheric accumulation of carbon dioxide to spatially resolved fluxes by utilizing the full suite of NASA data, models, and assimilation capabilities. For the oceanic part of this project, we introduce ECCO2-Darwin, a new ocean biogeochemistry general circulation model based on combining the following pre-existing components: (i) a full-depth, eddying, global-ocean configuration of the Massachusetts Institute of Technology general circulation model (MITgcm), (ii) an adjoint-method-based estimate of ocean circulation from the Estimating the Circulation and Climate of the Ocean, Phase II (ECCO2) project, (iii) the MIT ecosystem model “Darwin”, and (iv) a marine carbon chemistry model. Air–sea gas exchange coefficients and initial conditions of dissolved inorganic carbon, alkalinity, and oxygen are adjusted using a Green’s Functions approach in order to optimize modeled air–sea CO₂ fluxes. Data constraints include observations of carbon dioxide partial pressure (pCO₂) for 2009–2010, global air–sea CO₂ flux estimates, and the seasonal cycle of the Takahashi et al. (2009) Atlas. The model sensitivity experiments (or Green’s Functions) include simulations that start from different initial conditions as well as experiments that perturb air–sea gas exchange parameters and the ratio of particulate inorganic to organic carbon. The Green’s Functions approach yields a linear combination of these sensitivity experiments that minimizes model-data differences. The resulting initial conditions and gas exchange coefficients are then used to integrate the ECCO2-Darwin model forward. Despite the small number (six) of control parameters, the adjusted simulation is significantly closer to the data constraints (37% cost function reduction, i.e., reduction in the model-data difference, relative to the baseline simulation) and to independent observations (e.g., alkalinity). The adjusted air–sea gas exchange parameter differs by only 3% from the baseline value and has little impact (−0.1%) on the cost function. The particulate inorganic to organic carbon ratio was increased more than threefold and reduced the cost function by 22% relative to the baseline integration, indicating a significant influence of biology on air–sea gas exchange. The largest contribution to cost reduction (35%) comes from the adjustment of initial conditions. In addition to reducing biases relative to observations, the adjusted simulation exhibits smaller model drift than the baseline. We estimate drift by integrating the model with repeated 2009 atmospheric forcing for seven years and find a volume-weighted drift reduction of, for example, 12.5% for nitrate and 30% for oxygen in the top 300 m. Although there remain several regions with large model-data discrepancies, for example, overly strong carbon uptake in the Southern Ocean, the adjusted simulation is a first step towards a more accurate representation of the ocean carbon cycle at high spatial and temporal resolution.