Reduced-Complexity Models (RCMs), also known as Simple Climate Models (SCMs), are a highly parametrised variety of climate models designed to simulate Earth system dynamics in a computationally efficient manner. To achieve this, RCMs operate at low spatial and temporal resolutions, typically using global-mean annual averages, and represent many processes through parametrisation rather than explicit simulation. For instance, their temperature components commonly consist of zero-dimensional energy balance models that translate the imbalance between incoming and outgoing radiation into changes in global mean temperature. The nature and diversity of RCMs are discussed in greater detail by .

As a result of this efficiency, RCMs play an indispensable role in the climate science and policy landscape. By capturing the fundamental dynamics of the global climate system and carbon cycle with vastly greater computational speed than Earth System Models (ESMs), they enable systematic exploration of parametric and structural uncertainties across large numbers of climate simulations and the probabilistic quantification of uncertainty, tasks which are infeasible for ESMs alone. Further, the transparency and modularity of many RCMs allows for direct assessments of key physical and biogeochemical processes, thereby providing insights into the drivers of uncertainty. As such, RCMs enable the rapid integration of emerging scientific understanding, supporting real-time assessment of climate mitigation pathways, temperature goals and feedback sensitivities. Moreover, the absence of chaotic behaviour in these models enables the evaluation of climate impacts arising from comparatively small forcings that would otherwise be undistinguishable from interannual variability in ESMs. For instance, RCMs are a key tool in the calculation of the Social Cost from additional emissions of greenhouse gases GHGs,. These capabilities make them a vital accompaniment for process-resolving models for the assessment reports of the Intergovernmental Panel on Climate Change (IPCC).

RCMs have a long history of important scientific applications. Examples include assessing climate outcomes from Integrated Assessment Model (IAM)-derived emissions scenarios , direct coupling with IAMs , providing non-CO₂ adjustments to the remaining carbon budget , producing rapid answers in response to climate policy announcements , and assessing uncertainties induced by volcanic forcing on climate projections , to name a few. However, the trustworthiness of RCMs relies on their simulation skill. They need to be able to adequately emulate the behaviour of more complex climate models , to appropriately reproduce large-scale indicators of observed climate , and produce plausible projections of past and future climate change .

The Reduced-Complexity Model Intercomparison Project (RCMIP) was established to bring a coordinated and systematic approach to the use and analysis of these models. The first two phases of RCMIP focused on global responses to emissions scenarios. RCMIP Phase 1 provided a standardised protocol for generating the global-mean temperature projections for the IPCC's Sixth Assessment Report (AR6), demonstrating that the ensemble of RCMs was fit-for-purpose in reproducing historical warming and emulating the behaviour of the more complex models from CMIP5. This first protocol requested a single best-estimate ensemble member from each model provided, with the stipulation that this member should have an equilibrium climate sensitivity of 3 °C. RCMIP Phase 2 shifted focus from projection to evaluation, performing a deep diagnostic “stress test” of the probabilistic setups of various RCMs against a wide range of observational and ESM-derived benchmarks. A key finding was that no single model perfectly captured all assessed metrics, highlighting the importance of model diversity and the need to understand model-specific strengths and weaknesses.

These previous phases have provided critical lessons that motivate the design of RCMIP Phase 3 for the IPCC-AR7 cycle. Firstly, the model-specific calibration approaches used in Phase 1 made it difficult to disentangle the influence of model structural differences from calibration choices, limiting the robustness of structural uncertainty assessments. Further, an expanded evaluation beyond some key climate metrics such as global-mean temperature or effective radiative forcings was largely missing, limiting the ability to diagnose compensating errors and process-level biases. Moreover, with the scientific and policy focus demanding greater fidelity in understanding of net-zero emissions targets, a more rigorous evaluation of carbon cycle dynamics is required.

RCMIP Phase 3 (RCMIP3 hereafter) is therefore designed with two central goals: (1) to introduce a common, coordinated constraining protocol for model ensembles to enable a consistent assessment of inter-model structural differences, and (2) to focus the experimental design on key carbon cycle processes and climate indicators. This new phase is designed for maximum synergy with the upcoming Coupled Model Intercomparison Project Phase 7 (CMIP7) . The increasing focus of CMIP7 on emissions-driven simulations , including the new idealised experiments from flat10MIP which feature a constant CO₂ emissions rate of 10 Gt C yr⁻¹, provides a direct bridge to the native way of operating RCMs. RCMIP3 will leverage these synergies to provide rapid, probabilistic projections for CMIP7's “Assessment Fast Track” (AFT) experiments and to robustly assess key climate indicators like the Transient Climate Response to cumulative CO₂ Emissions (TCRE) and the Zero Emissions Commitment (ZEC). These indicators are central to the latest assessments of the state of the climate system .

Notwithstanding this focus on carbon cycle evaluation, the protocol has been designed alongside the community to transcend these objectives and support a wider evaluation of climate simulation by RCMs. The data request includes a comprehensive list of variables covering different aspects of RCM climate simulation, as well as thematic experiments focusing on specific climate factors such as methane emissions. As a result, we expect this protocol to enable multiple evaluation studies associated with this data request.

This paper outlines the full experimental protocol for RCMIP3, as illustrated in Fig. . The protocol is structured to systematically evaluate model behaviour and provide projections relevant to the IPCC AR7. In the following, we describe the experimental design including the new suite of idealised and non-idealised scenarios (Sect. ), the common constraining benchmarks (Sect. ), and the required output variables (Sect. ). Our goal is to enable a wide range of analyses based on this data request: carbon and methane cycle assessments, evaluations of key climate metrics such the TCRE and ZEC, and attributable warming studies.

The RCMIP3 experimental protocol has been compiled with broad community collaboration, following an open call for feedback on the draft protocol from RCM groups between August and October 2025. This allowed the co-design of the protocol with the wider RCM modelling community, incorporating input from representatives of multiple RCMs (ACC2, CICERO-SCM, FaIR, MAGICC, and OSCAR). The experiments are divided into two primary categories: idealised experiments designed to diagnose fundamental model properties, and non-idealised experiments that use comprehensive scenarios for historical simulation and future projection. The protocol is extensive, with 97 experiments in total, owing to the efficient nature of RCMs that enables this wider assessment. Notwithstanding, we have tiered our request, so that modelling teams can prioritise scenarios based on resource availability. The design and naming conventions of these experiments are consistent with those used in earlier MIPs where most of them were first formulated, thereby maintaining continuity across intercomparison projects and promoting consistency in interpretation. A spreadsheet with the full list of experiments, descriptions, durations and references is provided in the supplement materials (“scenario_info” sheet).

Input emissions, concentrations and forcing data (primarily natural forcing from solar cycles and volcanic eruptions) are provided for all experiments except those that need to be derived from model states (e.g., esm-1pct-brch-1000PgC). This input data includes an additional “spin-up” period of 1750–1850, which for idealised experiments is equal to piControl inputs, and for historical and scenario experiments are equal to historical inputs. Scenario durations are defined here as starting at the end of this spin-up period (see Tables  and ). All inputs include data for the full 1750–2500 period, but for experiments of a shorter duration, the period post experiment is denoted by missing values. For experiments where we recommend a longer run length (typically 1000 years) it is up to the discretion of the modellers how long they wish to extend the experiment. For all these, the last part is a continuation of values prior in the time series, recommended to look at slow-timescale dynamics, and extending the emissions or concentrations beyond what is provided should be trivial. If running the full 1000 years is prohibitive, a higher number of simulations for a shorter time period is preferable. For the most part, running the provided 751 years (including spin-up) will be sufficient. Reporting the output for the spin-up period is also not mandatory, and it is not necessary for models to run the spin-up period if it is not required for one or more experiments.

The emissions and concentrations data for experiments included in RCMIP phases I and II were taken from the datasets provided for those rounds . New data have either been generated according to stated protocols in the case of idealised datasets, or in the case of CMIP7, the inputs are from official CMIP7 sources including the CMIP7 Greenhouse Gas Concentrations , emissions and solar and volcanic forcing . An extensive overview of the CMIP7 forcing and data providers is found in . MethaneMIP data were generated according to the MethaneMIP protocol .

A central innovation of RCMIP3 is the introduction of a coordinated set of benchmarks for ensemble constraining and model evaluation. A key lesson from previous RCMIP phases was that model-specific calibration choices made it difficult to distinguish differences in projections arising from fundamental model structure versus those arising from the tuning process . To mitigate this problem, RCMIP3 includes a harmonised set of benchmarks all modelling teams are encouraged to use to constrain their ensemble output prior to submission, thereby fostering greater comparability and coordination across the RCM community. Additionally, these benchmarks will be used to evaluate model output once it is received from modelling teams, enabling a cleaner assessment of inter-model structural uncertainty, a key objective for the AR7 assessment cycle.

While providing this collection of climatic benchmarks, RCMIP3 does not prescribe to what extent the benchmarks are employed. This strategy allows participating groups to calibrate their models using their own preferred methods, while providing a common standardised experimental dataset that can be used where relevant. Ideally, all modelling groups incorporate these benchmarks in their tuning phases, but it is acknowledged that this may not be possible in all cases due to technical reasons or resource availability. Consequently, the use of these benchmarks is optional, but encouraged. Notwithstanding this flexibility, all benchmarks should be compared to simulation output for the CMIP7 esm-allGHG-historical scenario, including volcanic and solar forcing timeseries for CMIP7 as provided in the input datasets. Output from this scenario will also be used for model evaluation when comparing model submissions with the benchmarks.

These benchmarks are drawn from the latest observational datasets and community assessments, particularly the annual Indicators of Global Climate Change effort and the Global Carbon Budget . This ensures that the RCMIP3 process is anchored to the most up-to-date understanding of the Earth system. The number and nature of these targets have been carefully selected to encompass fundamental climate properties while minimising the burden on modelling groups.

To facilitate a consistent and efficient analysis of model results, a standardised data submission protocol is required. The format for RCMIP3 will broadly follow the structure used in previous phases with some modifications. Users should use the latest version of the pyrcmip tooling to validate and submit their model output, which has been modified to work with RCMIP3 and accepts both comma-separated value (CSV) as in previous phases and netCDF formats. A template for CSV formats for submitted time series data is provided in the supplementary material. The template also includes sheets for detailed model metadata and any regional definitions used, together with the possibility to add comments regarding the submitted model output. Jupyter notebooks providing an example on how to produce, validate and submit model output can be found in the notebooks folder in the pyrcmip tool referenced above. A complete list of all requested output variables, along with more detailed descriptions, tiers (see below), and units is provided in the supplementary material as well.

Our data request is divided into 3 main tiers [1–3], where tier 1 is the highest priority requested output which all groups are strongly encouraged to return to the extent they are capable. Subsequent lower tiers (2 and 3) cover domains less central to this RCMIP phase, supporting a more in-depth analysis. Additionally, there are also carbon (1-carbon) and methane (2-methane, 3-methane) specific tiers that are optionally requested from models that include the relevant processes. In this section, we describe the highest-priority variables, or tier 1, while categories entirely beyond tier 1 are listed in Table  (sea-level rise), Table  (emissions of minor species), Table  (methane cycle), and Table  (nitrogen cycle).

In this paper, we have presented the experimental design, constraining strategy, and data requirements for Phase 3 of the Reduced-Complexity Model Intercomparison Project (RCMIP3). This new phase is a significant evolution from its predecessors, designed to address key lessons learned and to provide a robust framework for the application of Reduced-Complexity Models in the IPCC AR7 assessment cycle and beyond.

The design philosophy of RCMIP3 is centred on two pillars: a coordinated and transparent approach to model evaluation and a special focus on carbon cycle dynamics. A major advancement from previous phases is the introduction of a common constraining strategy, encouraging all participating models to be benchmarked against a consistent set of observational and assessment-based targets. These targets are derived from the latest climate indicators and the Global Carbon Budget , and will allow for a much cleaner separation of uncertainty stemming from model structure versus calibration choices – a key ambiguity in past intercomparisons.

In thematic terms, this experimental design lays the foundation for the first comprehensive evaluation of RCM carbon cycle representations, a key component of Earth's climate system and a major limitation of past RCMIP phases. This focus on carbon cycle dynamics and metrics such as TCRE and ZEC is particularly timely, aligning directly with the scientific priorities of the AR7 cycle. Collectively, these design choices position RCMIP3 as a key tool to strengthen the integration across AR7 Working Groups I, II, and III, by providing a benchmarked and consistently calibrated ensemble of RCMs that supports a more robust assessment of climate outcomes for the full suite of emissions scenarios.

Furthermore, RCMIP3 is designed to provide substantial co-benefits for the wider climate modelling community, particularly in its synergy with CMIP7 . The efficiency of RCMs allows a comprehensive intercomparison protocol with 97 experiments, enabling a much broader set of results than would be available with an ESM ensemble of opportunity. The RCMIP3 protocol also aids in the exploration of different classes of model simulations, including all-GHG emissions-driven runs which are not yet a standard part of the CMIP protocol and methane-specific experiments. By running a broad suite of scenarios from current and previous CMIP phases, RCMIP3 provides a valuable link across assessment cycles. Furthermore, the experimental design significantly expands on CMIP7 idealised experiments like “flat10MIP” , using the computational efficiency of RCMs to explore sensitivities and overshoot dynamics in far greater detail than is feasible with comprehensive ESMs.

By establishing this common framework, RCMIP3 will not only provide critical data for the AR7 assessment but also deliver a lasting resource for the scientific community to better understand the behaviour of RCMs and their role in synthesising our knowledge of the Earth system.