This manuscript is likely to be of great interest to the paleoceanography community because it makes significant progress in finding a self-consistent decomposition of global mean benthic d18O into temperature and seawater (ice volume) components in a way which is consistent with independent estimates of global mean sea surface temperature (GMSST) and sea level constraints. Overall, it is well written and well supported by evidence. However, the manuscript could be significantly improved with some additional clarification.

Major points:

1. The calculations of mean ocean temperature (MOT) change relies on a transition in the ocean heat storage efficiency (HSE) from ~0.5 before the MPT to ~1 after the MPT. While the need for such a transition is well justified by comparison with BWT measurements, the available proxy data before the MPT (particularly in the Pacific) are quite sparse with large scatter and uncertainties. Although the authors appropriately provide a large uncertainty estimate for HSE, they provide calculations for the decomposition of d18Osw using only one scenario, in which HSE changes linearly between 1.5-0.9 Ma. It would be enormously helpful for the interpretation of the d18Osw estimate if the authors also provided the d18Osw results of a few sensitivity tests in which the timing and amplitude of HSE change are varied within the range consistent with BWT estimates.

2. Similarly, the timing of the hypothesized diagenetic alteration of benthic d18O is not well constrained by proxy data. Although Raymo et al (2018) proposed a simple linear trend for this effect, one might alternatively hypothesize that the effect would covary with MOT or BWT change if the mechanism responsible for the effect is the cooling of BWT. Because the manuscript estimates that MOT cools most dramatically during the MPT, it would be informative to also show the results of a sensitivity test in which the rate of diagenesis is greater for d18O immediately preceding the MPT (keeping the same estimated total diagenetic contribution at 3 Ma).

3. These two sensitivity tests would be particularly helpful for interpreting the unexpected observation that smoothed d18Osw and glacial maxima d18Osw at ~1.5 Ma are similar to (or possibly more enriched than) the d18Osw of the Late Pleistocene. It’s important to clarify whether this finding is relatively robust to the specified timing and amplitude of HSE change and d18O diagenesis, neither of which is well constrained by the available proxy data.

4. An additional surprising result is the relative amplitudes of orbital-scale MOT variability and orbital-scale d18Osw variability in the pre-MPT time period. The pre-MPT MOT record contains very weak glacial-interglacial change compared to relatively large amplitude d18Osw changes from 2.6-1.5 Ma. The authors should add some discussion of the reliability of the amplitudes of the orbital-scale signal in GMSST change and MOT change. Are the resolution and age uncertainty of the SST records sufficient to accurately estimate orbital-scale changes in GMSST and, thus, its application to estimating orbital responses in MOT and d18Osw?

5. In Figure 13, the authors present a very interesting comparison of BWT and d18Osw estimates from two Pacific cores and their global compilation estimates. They make the compelling argument that the estimates from the two cores are unlikely to provide reliable global estimates because they imply that sea level would need to be ~50 m higher than PI for significant amounts of time between 1.4-1 Ma, suggesting that these sites may be affected by local salinity changes. Could the authors slightly expand upon this idea to explain how the locations of those Pacific cores could have significantly different bottom properties than the rest of the deep Pacific?

6. I really appreciated the section of the paper using model results to explore the mechanisms responsible for scaling between MOT and GMSST and why it might differ before the MPT. However, one question I have is about the authors’ apparent conclusion that AABW’s contribution to MOT was constant (and approximately equal to pre-industrial) from 4-1.5 Ma (Figure 16D). How can this be consistent with the PlioMIP2 findings that the deep Southern Ocean was 1.5-2.5 C warmer than pre-industrial and that increased stratification caused decreased AABW formation?

Minor points:

Line 546: The statement that 1123 records large ice sheets pre-MPT is unclear because most of the pre-MPT d18Osw record is significantly lighter than the post-MPT glacial values. I think the authors might be referring to one particularly large glacial maximum at ~1.5 Ma. Please clarify exactly what is referred to here and how it provides support for the new d18Osw record.

Lines 605-606: The same text is repeated on these two lines.

Lines 647-648: The meaning of this sentence isn’t clear. Ice sheets have enhanced the warming relative to what? How is this visible in Figure 14C?

Figure 1: Many of the individual records are partially/mostly hidden behind other data in this figure. Also, the caption suggests that there are two different orange lines in the figure, which seems like a problem.

Figure 10B: It’s very hard to see the light blue line (which is an important result to be able to see) due to overlap with the gray line. Maybe make the shade of blue darker or leave off the gray line.

Figure 16F: The caption doesn’t provide the color information for all the different records shown.

This paper is a tour-de-force addressing difficult but highly relevant question of global climate evolution over the past 4.5 M years. This required the disentanglement of surface and deep ocean temperature under the constraint of global sea level rise that is derived from the benthic 18O stack. The analysis considered many other available high-resolution records covering this period and encompasses two apparently very different climatic regimes: the warm period before 1.5 Myr with smaller amplitude ice sheet and sea level cycles, primarily on the obliquity time scales, and after the Mid-Pleistocene Transition (MPT), i.e., post 1 M yr the familiar and well documented large-amplitude ice age cycles. The central result is that during the MPT the ocean heat storage efficiency (HSE) must have shifted from a low, constant value before, to twice that value, again constant, afterwards. This argument hinges on subtle differential changes between global mean SST and MOT that seem to be time dependent which leads to the hypothesis of a change in HSE.

The strength of this paper is the compilation and the comprehensive discussion of the wide paleoclimatic evidence from the various high-resolution archives. The weakness, however, is the motivation for the significant change in HSE and the underpinning modelling framework. Irrespective of this, the present contribution is important and forms part of a series of papers that address the structure and dynamics of global-scale climate during the last 4.5 M years. The authors should be encouraged to revise the paper, clarify the points raised in this review and provide a firmer and more robust modelling basis for their important conclusions.

Major comments:

1) A key conclusion is the HSE changes by a factor of 2 crossing the MPT. Hence, fundamentally the MPT is seen as a change in the ocean-atmosphere system through a combination of changes in sea ice cover, i.e. the transfer of heat from the atmosphere to the ocean, and changes in the repartitioning of heat between the surface/upper ocean and the deep ocean. The argument is motivated by an earlier study using a global comprehensive, isotope-enabled climate model (iCESM1) that simulated the last deglaciation (Zhu et al., 2024). It is not clear whether the different climate states that are visited in the 20 kyr-simulation encompass those that are relevant during the past 4.5 M years. I think the LGM-BA-YD-Hol sequence of the simulations is quite representative of the states that are visited after the MPT. However, I am not convinced that the same holds true for the climate states prior to 1.5 Myr ago. The reason is that generally higher global mean temperatures prevailed then with a quite different SST and polar temperatures. Overall, this was a situation of a substantially different ocean climate, particularly with respect to stratification which is the key process that may regulate HSE. The question is whether a comprehensive model under pre-MPT conditions would show a diminished HSE.

2) It appears that PlioMIP would be able to provide this underpinning. The model simulations presented in Weiffenbach et al 2024 are, unfortunately only analysing Southern Ocean processes, but surely the modelling results would be available to determine deltaSST, deltaMOT and all the quantities that are required to estimate HSE for that period. With such an analysis, all arguments in the present version to support a reduced HSE prior to the MPT could checked quantitatively. I am aware that this would be substantial additional work, but it would deliver the underpinning for the claims made in the paper.

3) Figures 14 and 15 should provide model-based insight for the HSE argument. However, it is not straightforward for the reader to connect the three cases (4x, mid Hol limited forcing, mid Hol full forcing) shown in Fig 14 with the conclusion of different HSE. Might the average of panel B be and approximation to HSE and (by visual estimate) about 1?? If so, the mean of panels D and F would also be approximately HSE with presumably HSE > 1??

Clarification of why you here show scaled ocean temperature changes would be appreciated.

The message of Fig 15 is not clear, and the figure could be possibly omitted.

4) Further to the modelling, the %-changes stated on line 690 seem quite fundamental to the argument, yet they pop up as a surprise and their derivation is not clear. I also do not find such analysis in Zhu et al 2024. Furthermore, how do these numbers connect with HSE? Where is 1.16 coming from, a few lines down? It seems that section 5 needs a major revision to make a convincing, model-based case.

Further comments:

5) Fig. 1, Caption: various shifts are mentioned. The shift by -1.73°C is motivated twice, first owing the entire 800 kyr, then to the Holocene. This is confusing.

6) Footnote 2: HSE is referring to equilibrium, HUE to transient changes. An important paper of reference is Zhu et al. 2024. In their figure 4, HSE is given as a function of time throughout the last 20 kyrs, in fact with longer periods where HSE>1, a case that is never discussed in this paper. A clarification of the different concepts and the relation to Zhu (used many times later in the paper) would be important.

7) Fig 2B: x-spread of red dots (LGM temperatures) is significantly less than purple dots (all temperatures of last 07. Myr) and yet the correlation is the same and the regression line goes far beyond the rightmost red dot (as far as I can recognize enlarging the figure)- I would expect a much reduced r^2 for the red data points.

8) Introduction: the intro is rather short, essentially only lines 43 to 62. From 63 to 121 a description of the current paper is given. Line 81, continued at 105 already provides the conclusion without having first given the overall context for HSE.

9) Fig 4 I: different time axis. Not sure whether this may be a mistake. Use the same time axis as in other panels; if there is no data prior to 3.4 Myr, then this should be left blank.

10) Fig 4 C: is there a cutoff in the red data at -2°C, but not in others?

11) entire paper: deltaMOT and deltaGMOT – what is the difference?

12) Fig 5B: refer again explicitly to Clark et al 2024 for the black line.

13) line 309: specify what long-term means.

14) line 317: this assertion is presumably based on visual inspection. If some quantitative approach is used, please specify.

15) line 324: clarify that the grey curve and cloud is based on HSE = 1 throughout the entire 4.5 Myr.

16) Fig 7: add labels of region to the panel, e.g. for A “eq Pacific”.

17) Fig 8 could be made more compelling by shifting downward the HSE curve so that it does not overlap with the two time series. Ad separate y-axis on the rhs, occupying much less vertical space for such a trivial curve (two straight lines linked by a slope).

18) Fig 9: the Mg/Ca constraint is really only operating in 1.5 to 3.2 Myr. This should be emphasized.

19) line 384: should this be 0 to +0.1 permil?

20) Fig 9, line 371: in text you use secular, here long-term. Please make consistent (btw secular would strictly mean century-scale (Latin saeculum), but likely you mean million-scale?)

21) line 449: should this rather be “increase” in d18O_T?

22) Fig 13: it would be helpful to have the ODP labels and locations at the right of the panel rows.

23) line 553: salinity changes are mentioned which is interesting. Could this be quantified. What would be the required magnitude? It this reasonable?

24) line 580: perhaps you add 30S to 30N and upper 500 m for the majority of heat uptake so the information follows the one on line 579.

25) line 583: Pacific warming is also smaller due to upwelling of colder intermediate and deep water. 

26) Fig 14 line 599: not clear which iTRACE simulations are meant. Once you use “full simulation” once just “simulation”.

27) line 605: duplication of sentence.

28) line 688: this is a very simple repartitioning model. It may serve the purpose, but the f’s and the deltaT’s should be diagnosed from model simulations. It is currently a bit obscure how these values (e.g., line 700, f=0.43) come about.