Paper check · Newton’s constant
NIST spent a decade rerunning BIPM’s torsion-balance measurement of Newton’s gravitational constant, G, on BIPM’s own instrument [1 Abstract][1 p. 27]. Its result sits 250 parts per million below BIPM’s and within the uncertainties of the consensus value. That says more about hidden laboratory errors than about gravity, though the cause hasn’t been found.
Our analysis marks our own reasoning and calculations, and the verdicts are our assessment. Everything else comes from the numbered sources.
6.67387(38)
NIST’s G, in 10−11 m3 kg−1 s−2; the (38) is the uncertainty in the last two digits, 57 parts per million [1 p. 27].
−250 ppm
Below BIPM’s corrected result from the same balance [1 Abstract].
−64 ppm
Below the CODATA 2022 value, within the uncertainties [1 p. 27].
The claims about the rerun
Schlamminger et al. 2026 [1 Abstract][1 p. 27]
It sits 250 parts per million below BIPM’s corrected 2014 value, about four standard uncertainties apart, and 64 below the CODATA 2022 value, about one, a difference the authors find not significant (p = 0.29). They add that their result “does not corroborate” the CODATA value either: it is consistent with it, not a confirmation. See section 1.
Schlamminger et al. 2026; the NIST release [1 p. 3][2]
One instrument gave answers 250 parts per million apart in two laboratories. The authors call new physics “remote” and hidden instrumental effects “more plausible”, but they found no single cause, and the test masses, the disk, the metrology and the analysis all changed between the two runs [1 p. 9][1 p. 28]. See section 5.
“The team found virtually identical results.”
The NIST release, on copper and sapphire source masses [2]
By our arithmetic from the paper’s tables, the sapphire runs sit 150 parts per million below the copper runs with the servo method, about four times their combined uncertainty, and 58 below with free deflection, well within it; and the paper finds “larger unexplained effects in the sapphire measurements” [1 p. 25][1 p. 26][1 p. 27]. See section 3.
“the new result differs from the standard by 0.0235%”
Universe Magazine’s summary of its April 21, 2026 article [3]
The release’s 0.0235% is the gap between NIST’s and BIPM’s values [2]. NIST’s result is 0.0064% below the CODATA value, a difference the authors find not significant [1 p. 27]. The article’s own text comes closer, giving the figure as the difference from “the previous reference measurement” [3]. See section 5.
1 · Two laboratories
BIPM’s torsion balance measured the gravitational constant, G, in Sèvres, France, in 2007. Shipped to NIST in Gaithersburg, Maryland, in 2016, the same apparatus, with new test masses, a new disk and new metrology, gave a value 250 parts per million lower [1 p. 4][1 p. 5][1 p. 9].
Difference from the CODATA 2022 value, parts per million (bars: ±1 standard uncertainty)
Their value is “inconsistent with the earlier determination of G obtained using the same apparatus”, while its difference from the CODATA value “does not differ significantly from zero (p-value = 0.29)”. Yet “Our result does not corroborate either the earlier BIPM value or the current CODATA recommendation” [1 p. 27].
A 64 ppm gap with a 57 ppm uncertainty is consistent with the CODATA value, but it is not a precise confirmation of it. NIST’s uncertainty is more than twice CODATA’s, and the final value combines four sub-results that disagree with one another, which the authors handle by adding “dark uncertainty” and by using a prior centred on the CODATA value (sections 3 and 5).
2 · Four decades
CODATA’s value is a weighted mean of 16 measurements reported from 1982 to 2018. They scatter so much more than their uncertainties allow that CODATA multiplies every uncertainty by 3.9 before averaging [4 p. 025002-41].
NIST-82 6.67248(43)
TR&D-96 6.6729(5)
LANL-97 6.67398(70)
UWash-00 6.674255(92)
BIPM-01 6.67559(27)
UWup-02 6.67422(98)
MSL-03 6.67387(27)
HUST-05 6.67222(87)
UZur-06 6.67425(12)
HUST-09 6.67349(18)
BIPM-14 6.67554(16)
LENS-14 6.67191(99)
UCI-14 6.67435(13)
HUSTT-18 6.674184(78)
HUSTA-18 6.674484(77)
JILA-18 6.67260(25)
NIST-26 6.67387(38)
Spread
Our analysis≈ 550 ppm
From LENS-14, the lowest of the 16, to BIPM-01, the highest [4 Table XXX].
Scatter
4.2×
The standard deviation of the values is 4.2 times their median reported uncertainty [6 p. 4].
Expansion factor
3.9×
CODATA multiplies every input uncertainty by 3.9, which brings each value within two of its expanded uncertainties of the mean [4 p. 025002-41].
3 · Tests
Four source masses pull on four test masses on a disk that hangs from a thin strip; the twist, or the voltage needed to stop it, gives G [1 p. 4][1 p. 5].
In free deflection the disk twists and the angle is measured, as in Cavendish’s method; in the servo method, voltages on the electrodes hold it still. Each gives its own estimate of G [1 p. 4][1 p. 7].
Bryan Kibble had suggested that magnetic coupling between the masses inflated BIPM’s value; a 2023 study put an upper limit of about 20 ppm on that effect, and NIST added source masses of single-crystal sapphire “to test this conclusively” [1 p. 9][7]. The sapphire results came out 150 ppm (servo, about 4σ) and 58 ppm (free deflection, within the uncertainties) below copper, by our arithmetic from the paper’s tables, and the paper flags “larger unexplained effects” in them [1 p. 25][1 p. 26][1 p. 27]. It doesn’t say the test settled Kibble’s idea.
Diamond-turned test masses pin down where each mass sits; a new disk replaced the original, which had slightly non-parallel faces [1 p. 9].
NIST’s mass group multiplied every source-mass value by a secret factor within ±0.1% and sealed it in an envelope, to prevent experimenter bias [1 p. 27]. The release says the worry was drifting toward BIPM’s value, and describes Patrick Abbott subtracting a number from the weights of some of the masses [2].
The team found a spurious torque that a temperature difference of just five millionths of a degree over 10 cm inside the vacuum can could produce through the little gas left inside, one of several effects “not previously treated for this apparatus” [1 p. 19][1 p. 28]. At BIPM’s likely pressure the authors put it at no more than about 10 ppm, too small to explain the gap, but it may matter in balances with smaller signals [1 p. 28].
4 · History
BIPM measures G with the rebuilt Mark II balance in Sèvres, France [1 p. 4].
BIPM publishes 6.67545(18). A 2014 erratum, after a source-mass density correction had been applied twice, gives 6.67554(16) [8][6 p. 4].
After BIPM’s Terry Quinn proposed it at a 2014 NIST workshop, the balance is shipped to NIST for an independent team to rerun [1 p. 3][1 p. 5].
A planned unblinding is called off when the team spots an overlooked effect “related to air pressure” [2].
The envelope is opened during a talk at the CPEM conference in Aurora, Colorado. The secret offset is large and negative, but too large for the result to match BIPM’s [2][1 p. 27].
After about two more years of analysis, the result appears in Metrologia: 6.67387(38). The value shown at the unblinding was 44 ppm lower; later measurements of where the test masses sit moved it [1 p. 27].
5 · Read with care
This section is our assessment, checked against the paper, the release, the CODATA tables and the coverage.
One outlet’s summary says the result differs from the standard by 0.0235% [3]; its article text attributes the figure to “the previous reference measurement”. That figure is the release’s gap between NIST and BIPM [2]. Against the CODATA value the gap is 64 ppm, 0.0064%, and not significant [1 p. 27].
The paper’s abstract and Figure 27 compare NIST with BIPM’s corrected 2014 value: 2.5 × 10−4 [1 Abstract][1 p. 27]. Its closing section instead places BIPM 1.71 × 10−4 above CODATA and NIST 0.64 × 10−4 below, which adds up to the release’s 0.0235% [1 p. 27][2]. By our arithmetic the corrected BIPM value is 1.86 × 10−4 above CODATA; 1.71 is closer to BIPM’s superseded 2013 value, 1.72 above.
NIST’s publications database still lists 6.67366 ± 0.00020, a relative uncertainty of “3.0 × 10−6” (that ratio is 3.0 × 10−5) and a 2.81 × 10−4 gap to BIPM, apparently from an earlier version [9]. The paper gives 6.67387(38); the release, 6.67387 [1 p. 27][2].
The blind protected the analysis until July 2024. The value shown then was 44 ppm lower than the published one, which the authors attribute to later measurements of the test masses [1 p. 27]. The final uncertainty also includes “dark uncertainty” for the scatter among the four sub-results, and the combination uses a prior centred on the CODATA value, with a standard deviation of about 135 ppm [1 p. 26].
One instrument giving two answers in two laboratories points to errors in how the measurement was made or analysed, at BIPM, at NIST or both. The authors quantified several effects not previously treated for this apparatus but found no single dominant cause, “leaving the larger discrepancy unresolved” [1 p. 28].
Clive Speake contributed to both the BIPM and NIST efforts, so the authors call their determination “not entirely independent” [1 p. 3].
Bottom line: a meticulous rerun that is consistent with the consensus value and disagrees with the instrument’s own earlier answer.
The original infographic’s reading holds: the 250 ppm gap says more about hidden laboratory errors than about gravity. But copper and sapphire didn’t give “virtually identical” results, the sapphire test didn’t settle Kibble’s idea, and the cause of the BIPM discrepancy is still unknown [1 p. 27][1 p. 28].
Bracketed numbers refer to this list; p. gives a page (for the journal articles, the journal’s page numbers), Table a table, and Abstract the part of a work we used. Links checked October 14, 2026.
Schlamminger, S., Chao, L., Lee, V., Shakarji, C., Possolo, A., Newell, D., Stirling, J., Cochrane, R., and Speake, C. (2026). Redetermination of the gravitational constant with the BIPM torsion balance at NIST. Metrologia 63(2), 025012. Published April 16, 2026. CC BY 4.0.
National Institute of Standards and Technology (2026). NIST Weighs In on the Mystery of the Gravitational Constant. News release, April 16, 2026 (page updated May 5, 2026).
Nehriienko, R. (2026). Problem G: New precise measurements of the gravitational constant disagree with previous ones again. Universe Magazine, April 21, 2026; the page’s metadata names Liliia Alaieva. The quotation is from the article’s search and link-preview summary (its meta description); its text says the new value “differs from the previous reference measurement by 0.0235%”.
Mohr, P. J., Newell, D. B., Taylor, B. N., and Tiesinga, E. (2025). CODATA recommended values of the fundamental physical constants: 2022. Reviews of Modern Physics 97(2), 025002. Table XXX lists the 16 input values for G; we read the copy NIST hosts.
Rosi, G., Sorrentino, F., Cacciapuoti, L., Prevedelli, M., and Tino, G. M. (2014). Precision measurement of the Newtonian gravitational constant using cold atoms. Nature 510, 518–521. We read the accepted manuscript, arXiv:1412.7954.
Merkatas, C., Toman, B., Possolo, A., and Schlamminger, S. (2019). Shades of dark uncertainty and consensus value for the Newtonian constant of gravitation. Metrologia 56(5), 054001. CC BY 3.0.
Speake, C. C., Bryant, J. L., Davis, R. S., and Quinn, T. J. (2023). The influence of time varying magnetic fields on the determinations of Newton’s constant at the Bureau International des Poids et Mesures. Metrologia 60(2), 024001. CC BY 4.0.
Quinn, T., Parks, H., Speake, C., and Davis, R. (2013). Improved determination of G using two methods. Physical Review Letters 111, 101102; erratum, Physical Review Letters 113, 039901 (2014). We read the 2013 abstract; the erratum’s content is as Merkatas and colleagues and CODATA report it.
National Institute of Standards and Technology. Redetermination of the gravitational constant with the BIPM torsion balance at NIST. NIST publications database entry (updated May 2, 2026), read October 14, 2026.
Method
On October 14, 2026, we checked every statement, number and source on the original infographic, 60 items, against the published paper, the NIST release, the CODATA 2022 tables, the 2019 analysis of the CODATA inputs and the papers they cite. We recomputed every difference, ratio and σ value and measured the original charts against their axes. The check corrected several statements; the change log lists them.
After the page was built, a second reviewer who hadn’t seen the drafting recomputed every number and chart mark from the sources, which matched, and checked every quotation and reference. The changes that followed: the Universe Magazine article is credited to its visible author, and the quoted phrase is identified as its search and link-preview summary, while its text gives the 0.0235% as the difference from “the previous reference measurement”; the copper–sapphire comparison gives the uncertainties (the servo results differ by about four standard uncertainties, the free-deflection results agree within them); the temperature difference behind the gas torque is over 10 cm inside the vacuum can, and the authors’ 10 ppm is an upper bound; the blind’s purpose is the paper’s, to prevent experimenter bias; the dark uncertainty and the prior appear where the page discusses agreement with CODATA; and several locators were added.
The verdicts and the items tagged “Our analysis” are our own assessment, built on the cited sources: the σ comparisons, the spread of the 16 values, the copper–sapphire differences and our reading of the release’s 0.0235%. The paper is open access under CC BY 4.0 [1]; the charts are our own, drawn from the CODATA input values and the paper’s result.
This page was researched and written with Claude, an AI assistant made by Anthropic, and then re-checked against its sources in a separate verification pass.
Corrections
If we get something wrong, we fix it and record the change below with the date. If a correction changes a verdict, the log says what the old verdict was.