COSMOS2025: Definitive JWST COSMOS-Web Catalog

• COSMOS2025 is a multi-wavelength catalog that combines deep JWST/NIRCam and MIRI imaging with 37-band ground-based data to precisely measure over 700,000 galaxies.
• The catalog employs dual photometric strategies—fixed-aperture and profile fitting—to deliver consistent and accurate structural, spectral, and redshift estimates.
• COSMOS2025 provides a robust resource for studying galaxy evolution, environmental effects, protoclusters, and high-redshift populations in the JWST era.

Searching arXiv for COSMOS2025 and closely related papers to ground the article in current literature. arXiv search query: COSMOS2025 catalog COSMOS-Web photometry morphology redshifts physical parameters COSMOS2025 is the COSMOS-Web catalog of photometry, morphology, photometric redshifts, and physical parameters for more than 700,000 galaxies in the Cosmic Evolution Survey field, built on the \textit{James Webb Space Telescope} 255\,h COSMOS-Web program and designed as the definitive JWST+legacy multi-wavelength catalog in the central COSMOS area (Shuntov et al., 3 Jun 2025). It combines deep NIRCam imaging over the central $\sim 0.54\,{\rm deg}^2$ and parallel MIRI imaging over $\sim 0.2\,{\rm deg}^2$ with 32 ancillary bands from the observed UV to the mid-IR, and provides matched photometry, structural measurements, machine-learning morphology, photometric-redshift PDFs, and SED-based physical properties that have already been used for studies of protoclusters, galaxy groups, dusty star-forming galaxies, stellar mass functions, and morphology out to high redshift (Shuntov et al., 3 Jun 2025).

1. Survey definition and observational scope

COSMOS2025 is constructed to support the JWST-led COSMOS-Web Survey and to extend the legacy of earlier COSMOS releases, including COSMOS2009, COSMOS2015, COSMOS2020, and COSMOS2020 (Toni et al., 9 Sep 2025). The parent imaging combines JWST/NIRCam F115W, F150W, F277W, and F444W, JWST/MIRI F770W, HST/ACS F814W, CFHT/MegaCam $u^\star$, Subaru/HSC optical and narrow-band imaging, Subaru/Suprime-Cam intermediate- and narrow-band imaging, UltraVISTA near-infrared imaging, and Spitzer/IRAC channels, yielding 37-band photometry spanning $0.3$–$8\,\mu$m (Shuntov et al., 3 Jun 2025).

The survey geometry is defined by a central NIRCam footprint of $0.54\,{\rm deg}^2$ and a MIRI footprint of $0.20\,{\rm deg}^2$; the ground-based data cover a wider $\sim 2\,{\rm deg}^2$ COSMOS field, but only the central $0.54\,{\rm deg}^2$ is used in COSMOS2025 (Shuntov et al., 3 Jun 2025). The COSMOS-Web morphological catalog reports $\sim 780{,}000$ galaxies, while the combined four-band NIRCam detection image yields 784,016 sources in the parent catalog (Yang et al., 12 Jun 2026). In the stellar-mass-function analysis, the effective area after bright-star masking is $\sim 0.2\,{\rm deg}^2$0 ($\sim 0.2\,{\rm deg}^2$1), and the catalog contains $\sim 0.2\,{\rm deg}^2$2 galaxies to $\sim 0.2\,{\rm deg}^2$3 (Shuntov et al., 7 Nov 2025).

A compact summary of the principal observational components is given below.

Component	Coverage	Key specification
NIRCam	$\sim 0.2\,{\rm deg}^2$4	F115W, F150W, F277W, F444W
MIRI	$\sim 0.2\,{\rm deg}^2$5	F770W
Photometric span	central COSMOS footprint	37 bands over $\sim 0.2\,{\rm deg}^2$6–$\sim 0.2\,{\rm deg}^2$7m

The nominal $\sim 0.2\,{\rm deg}^2$8 empty-aperture depths in the JWST bands are F115W$\sim 0.2\,{\rm deg}^2$9, F150W$u^\star$0, F277W$u^\star$1, F444W$u^\star$2, and F770W$u^\star$3 AB mag (Shuntov et al., 3 Jun 2025). A related description of the COSMOS-Web morphology release quotes NIRCam depths of $u^\star$4–$u^\star$5 AB mag in $u^\star$6 apertures and MIRI depths of $u^\star$7–$u^\star$8 AB mag in $u^\star$9 apertures, reflecting the measurement configuration used in that catalog (Yang et al., 12 Jun 2026).

2. Catalog construction, photometry, and source modeling

COSMOS2025 provides two complementary photometric products for JWST-detected sources (Shuntov et al., 3 Jun 2025). The first is a “hot+cold” fixed-aperture catalog on PSF-homogenized space-based bands. Detection is performed on a $0.3$0 image built from the four NIRCam SNR maps; cold-mode extraction uses a $0.3$1 threshold, a large top-hat filter, $0.3$2 px, and $0.3$3, while hot-mode extraction uses a $0.3$4 threshold, a small Gaussian filter, $0.3$5 px, and $0.3$6 (Shuntov et al., 3 Jun 2025). Aperture photometry is measured in circular apertures of $0.3$7, $0.3$8, $0.3$9, $8\,\mu$0, and $8\,\mu$1, together with Kron elliptical apertures, and corrected to “total” using the ratio to $8\,\mu$2 Kron on F444W plus PSF corrections (Shuntov et al., 3 Jun 2025).

The second product is a SourceXtractor++ full 37-band profile-fitting catalog on native-resolution images (Shuntov et al., 3 Jun 2025). PSF models are built per tile and band with PSFEx, overlapping sources are grouped through Kron-based ellipses and UltraVISTA $8\,\mu$3 segmentation, and fitting proceeds through iterative “meta-iterations” with sequential subtraction (Shuntov et al., 3 Jun 2025). The modeling run on the NIRCam bands fits either a single Sérsic model with parameters $8\,\mu$4 or a bulge+disk model with fixed $8\,\mu$5 bulge and $8\,\mu$6 disk, followed by forced photometry in the remaining 33 bands with structural parameters fixed (Shuntov et al., 3 Jun 2025).

Noise and uncertainty calibration are empirical rather than purely photon-noise based. In the hot+cold catalog, $8\,\mu$7 random apertures are placed and the negative tail is fit to derive

$8\,\mu$8

where $8\,\mu$9 is the aperture area (Shuntov et al., 3 Jun 2025). In the SE++ catalog, covariance-matrix errors are reported to be underestimated by $0.54\,{\rm deg}^2$0, so an effective aperture area $0.54\,{\rm deg}^2$1 is defined from the ellipse enclosing 90\% of the convolved model flux, and a background term $0.54\,{\rm deg}^2$2 is added in quadrature (Shuntov et al., 3 Jun 2025).

This dual strategy is central to the catalog’s design. Fixed-aperture measurements provide homogeneous space-based photometry, while profile fitting extends consistent flux estimation across the full 37-band set without PSF homogenization (Shuntov et al., 3 Jun 2025). The combination is a defining feature of COSMOS2025 rather than a secondary post-processing layer.

3. Redshifts, SED fitting, and completeness

Photometric redshifts and rest-frame quantities are derived primarily with LePHARE, using a BC03 template library, six star-formation histories, two metallicities, 43 ages from $0.54\,{\rm deg}^2$3–$0.54\,{\rm deg}^2$4 Gyr, emission lines, dust attenuation curves from Calzetti 2000, Arnouts 2013, and Salim 2018, dust emission templates from Béthermin 2012 and Magdis 2012, and IGM absorption following Madau 1995 (Shuntov et al., 3 Jun 2025). The redshift PDF is formed from $0.54\,{\rm deg}^2$5 summed over templates at each redshift, with $0.54\,{\rm deg}^2$6 defined by the PDF median and a 68\% confidence interval also reported (Shuntov et al., 3 Jun 2025).

Against a spectroscopic sample of $0.54\,{\rm deg}^2$7 high-confidence redshifts out to $0.54\,{\rm deg}^2$8, the catalog achieves $0.54\,{\rm deg}^2$9 at $0.20\,{\rm deg}^2$0, with outlier rate $0.20\,{\rm deg}^2$1 and bias $0.20\,{\rm deg}^2$2 (Shuntov et al., 3 Jun 2025). The same paper reports $0.20\,{\rm deg}^2$3 as a function of magnitude, color, and galaxy type, and describes this as a factor of $0.20\,{\rm deg}^2$4 improvement at 26 AB mag compared to COSMOS2020 (Shuntov et al., 3 Jun 2025). An independent COSMOS2025 group analysis quotes $0.20\,{\rm deg}^2$5 with $0.20\,{\rm deg}^2$6 catastrophic failures for bright galaxies and $0.20\,{\rm deg}^2$7 with $0.20\,{\rm deg}^2$8 outliers for $0.20\,{\rm deg}^2$9 up to 28 (Toni et al., 9 Sep 2025). In the protocluster analysis, the quoted precision is

$\sim 2\,{\rm deg}^2$0

for sources with $\sim 2\,{\rm deg}^2$1, valid out to $\sim 2\,{\rm deg}^2$2 (Li et al., 15 May 2026).

Physical parameters are produced both by LePHARE and by CIGALE (Shuntov et al., 3 Jun 2025). The LePHARE “z-fixed” run returns $\sim 2\,{\rm deg}^2$3, SFR, age, $\sim 2\,{\rm deg}^2$4, and $\sim 2\,{\rm deg}^2$5 at the median $\sim 2\,{\rm deg}^2$6, while CIGALE employs the \texttt{sfhNlevels} non-parametric SFH module with a continuity-burst prior to estimate stellar mass, $\sim 2\,{\rm deg}^2$7, dust attenuation, metallicity, ages such as $\sim 2\,{\rm deg}^2$8, and a migration vector on the SFR–$\sim 2\,{\rm deg}^2$9 plane (Shuntov et al., 3 Jun 2025). Validation against mock catalogs and the Horizon-AGN simulation gives mass bias $0.54\,{\rm deg}^2$0 dex and SFR bias $0.54\,{\rm deg}^2$1 dex for the CIGALE run (Shuntov et al., 3 Jun 2025).

Completeness is quantified with the Pozzetti et al. method. The rescaled mass is defined as

$0.54\,{\rm deg}^2$2

with $0.54\,{\rm deg}^2$3 in the main catalog completeness calculation (Shuntov et al., 3 Jun 2025). COSMOS2025 is described as approximately 80\% complete at $0.54\,{\rm deg}^2$4 at $0.54\,{\rm deg}^2$5 and at $0.54\,{\rm deg}^2$6 at $0.54\,{\rm deg}^2$7, representing a gain of $0.54\,{\rm deg}^2$8 dex compared to COSMOS2020 (Shuntov et al., 3 Jun 2025). This depth is one reason why the catalog supports both high-redshift searches and low-mass galaxy population studies in a uniform framework.

4. Morphology, structural measurements, and classification

Morphology is a core COSMOS2025 product rather than a derivative add-on. The catalog provides parametric measurements from SE++, independent measurements from Galight/Lenstronomy, non-parametric diagnostics such as Gini and $0.54\,{\rm deg}^2$9, and machine-learning classification probabilities in each NIRCam band (Shuntov et al., 3 Jun 2025). For all NIRCam-detected sources with $\sim 780{,}000$0, the single-Sérsic $\sim 780{,}000$1 distribution peaks near $\sim 780{,}000$2, extends to $\sim 780{,}000$3, and the effective-radius distribution peaks near $\sim 780{,}000$4; the bulge+disk $\sim 780{,}000$5 distribution peaks near $\sim 780{,}000$6 for disks and $\sim 780{,}000$7 for spheroids (Shuntov et al., 3 Jun 2025).

The dedicated COSMOS-Web morphology release extends these measurements to $\sim 780{,}000$8 galaxies and performs both parametric and non-parametric morphology analyses independently in the four NIRCam bands (Yang et al., 12 Jun 2026). Single-Sérsic fits adopt

$\sim 780{,}000$9

with bounds $\sim 0.2\,{\rm deg}^2$00, $\sim 0.2\,{\rm deg}^2$01, and $\sim 0.2\,{\rm deg}^2$02 (Yang et al., 12 Jun 2026). Bulge–disk decompositions use two fixed-$\sim 0.2\,{\rm deg}^2$03 components, with $\sim 0.2\,{\rm deg}^2$04 and additional constraints that the bulge effective radius be smaller than the disk effective radius, the centroids agree within 1 px, and the disk be more elliptical (Yang et al., 12 Jun 2026).

The morphology catalog reports a strong correlation between structure and star-formation activity up to $\sim 0.2\,{\rm deg}^2$05 (Yang et al., 12 Jun 2026). On the SFR–$\sim 0.2\,{\rm deg}^2$06 plane, exponential profiles with $\sim 0.2\,{\rm deg}^2$07 trace the star-forming main sequence, while quiescent below-main-sequence galaxies have $\sim 0.2\,{\rm deg}^2$08 and $\sim 0.2\,{\rm deg}^2$09; at $\sim 0.2\,{\rm deg}^2$10 this bimodality weakens and many quiescent systems remain disk-like with $\sim 0.2\,{\rm deg}^2$11 (Yang et al., 12 Jun 2026). The redshift evolution is strongly mass dependent: for $\sim 0.2\,{\rm deg}^2$12, the median $\sim 0.2\,{\rm deg}^2$13 increases from $\sim 0.2\,{\rm deg}^2$14 at $\sim 0.2\,{\rm deg}^2$15 to $\sim 0.2\,{\rm deg}^2$16 at $\sim 0.2\,{\rm deg}^2$17, whereas lower-mass galaxies remain at $\sim 0.2\,{\rm deg}^2$18 at all epochs (Yang et al., 12 Jun 2026). In the same mass range, optical $\sim 0.2\,{\rm deg}^2$19 rises from $\sim 0.2\,{\rm deg}^2$20 at $\sim 0.2\,{\rm deg}^2$21 to $\sim 0.2\,{\rm deg}^2$22 at $\sim 0.2\,{\rm deg}^2$23 (Yang et al., 12 Jun 2026).

A separate machine-learning study develops a fast, interpretable classifier for early- and late-type galaxies at $\sim 0.2\,{\rm deg}^2$24 using 66 broadband colors and a CatBoostClassifier trained on a simulation-based set of secure labels (Asadi et al., 2 Jun 2026). In the held-out simulated test set, late types are recovered with 97.7\% precision and 98.3\% recall, while early types reach 90.8\% precision and 88.1\% recall; applied to 44,132 COSMOS2025 galaxies, only 5.9\% fall in the ambiguous range $\sim 0.2\,{\rm deg}^2$25 (Asadi et al., 2 Jun 2026). The most important feature is F277W$\sim 0.2\,{\rm deg}^2$26F444W, which the paper interprets as tracing the optical/NIR contrast between old and young stellar populations (Asadi et al., 2 Jun 2026). This suggests that broadband SED shape, not only direct image morphology, is already highly informative for large-scale structural classification in COSMOS2025.

5. Environmental structure, overdensities, and the build-up of passive populations

COSMOS2025 has been used to identify galaxy groups with AMICO, a linear matched-filter algorithm operating in the 3D space of sky position and photometric redshift and using an NFW projected radial profile, a Schechter luminosity function in the F150W reference band, and the photometric-redshift PDF of each galaxy (Toni et al., 9 Sep 2025). The algorithm computes an amplitude

$\sim 0.2\,{\rm deg}^2$27

and records peaks above the threshold $\sim 0.2\,{\rm deg}^2$28 as group candidates (Toni et al., 9 Sep 2025). Membership probabilities are then assigned through

$\sim 0.2\,{\rm deg}^2$29

This selection is explicitly described as unbiased with respect to the presence or absence of a red sequence (Toni et al., 9 Sep 2025).

Within these groups, quiescent fractions are estimated both through membership weighting,

$\sim 0.2\,{\rm deg}^2$30

and through a model-independent cylinder background-subtraction method (Toni et al., 9 Sep 2025). Across all richness bins, $\sim 0.2\,{\rm deg}^2$31 remains very low ($\sim 0.2\,{\rm deg}^2$32) at $\sim 0.2\,{\rm deg}^2$33, then rises steeply for $\sim 0.2\,{\rm deg}^2$34, reaching $\sim 0.2\,{\rm deg}^2$35 by $\sim 0.2\,{\rm deg}^2$36 in the richest systems with $\sim 0.2\,{\rm deg}^2$37 (Toni et al., 9 Sep 2025). The first galaxies settle onto the red-sequence ridgeline by $\sim 0.2\,{\rm deg}^2$38, and a compact overdensity at $\sim 0.2\,{\rm deg}^2$39 with $\sim 0.2\,{\rm deg}^2$40 and five spectroscopic members is reported as a rare early red sequence (Toni et al., 9 Sep 2025). The same study finds that X-ray faint groups have, on average, lower quiescent fractions than X-ray bright ones (Toni et al., 9 Sep 2025).

At higher redshift, COSMOS2025 underpins the identification of the protocluster PC J1001+0214 at $\sim 0.2\,{\rm deg}^2$41 in the COSMOS field (Li et al., 15 May 2026). Using the COSMOS-3D WFSS footprint and strict photometric-redshift quality cuts, the analysis defines photometric members through a re-run of LePHARE on a fine $\sim 0.2\,{\rm deg}^2$42 grid and obtains 131 total members: 21 spectroscopic and 110 photometric (Li et al., 15 May 2026). The overdensity statistic is computed from corrected aperture counts,

$\sim 0.2\,{\rm deg}^2$43

and

$\sim 0.2\,{\rm deg}^2$44

In the $\sim 0.2\,{\rm deg}^2$45–3.1 slice, the structure reaches $\sim 0.2\,{\rm deg}^2$46, placing it in the top 0.51\% of all apertures (Li et al., 15 May 2026).

The galaxy population in this protocluster shows a Gaussian-fit peak in stellar mass at $\sim 0.2\,{\rm deg}^2$47, compared to $\sim 0.2\,{\rm deg}^2$48 in the coeval field, corresponding to $\sim 0.2\,{\rm deg}^2$49 dex (Li et al., 15 May 2026). Median SFR offsets relative to the field are +0.117 dex for $\sim 0.2\,{\rm deg}^2$50–8.6, +0.147 dex for 9.0–9.4, +0.115 dex for 9.4–9.7, and $\sim 0.2\,{\rm deg}^2$51 dex for 8.6–9.0 (Li et al., 15 May 2026). The quiescent fraction at $\sim 0.2\,{\rm deg}^2$52 is $\sim 0.2\,{\rm deg}^2$53 in the protocluster and $\sim 0.2\,{\rm deg}^2$54 in the field, leading the authors to conclude that environmental quenching has not yet become dominant (Li et al., 15 May 2026). A plausible implication is that COSMOS2025 is sufficiently deep and internally homogeneous to detect environmental effects at a stage when mass assembly has accelerated but strong quenching has not.

6. High-redshift dusty galaxies, stellar mass functions, and catalog legacy

COSMOS2025 also supports searches for populations that are difficult to isolate in conventional optical or submillimeter selections. One example is the identification of faint dusty star-forming galaxies at $\sim 0.2\,{\rm deg}^2$55 by combining ALMA CHAMPS sources with COSMOS2025 photometry (Zavala et al., 18 Dec 2025). That analysis defines a “stellar index”

$\sim 0.2\,{\rm deg}^2$56

and applies the cuts $\sim 0.2\,{\rm deg}^2$57, $\sim 0.2\,{\rm deg}^2$58 mag, and $\sim 0.2\,{\rm deg}^2$59, with a size-based rejection criterion $\sim 0.2\,{\rm deg}^2$60 to remove point sources (Zavala et al., 18 Dec 2025). Within the $\sim 0.2\,{\rm deg}^2$61 CHAMPS area, 20 candidates are selected and 18 survive visual inspection (Zavala et al., 18 Dec 2025). Inverse-variance weighted stacking of the individually undetected 1.2 mm maps yields a $\sim 0.2\,{\rm deg}^2$62 detection with

$\sim 0.2\,{\rm deg}^2$63

from the direct rms and

$\sim 0.2\,{\rm deg}^2$64

from the bootstrap rms (Zavala et al., 18 Dec 2025). The corresponding surface density is $\sim 0.2\,{\rm deg}^2$65 and the comoving volume density is $\sim 0.2\,{\rm deg}^2$66 over $\sim 0.2\,{\rm deg}^2$67 (Zavala et al., 18 Dec 2025).

The catalog is also the basis for a morphology-dependent stellar-mass-function analysis over ten redshift bins at $\sim 0.2\,{\rm deg}^2$68 (Shuntov et al., 7 Nov 2025). Quiescent galaxies are selected through the Ilbert et al. rest-frame criterion

$\sim 0.2\,{\rm deg}^2$69

yielding 15,608 quiescent and 268,394 star-forming systems in a final sample of 284,002 galaxies after removal of stars, brown dwarfs, and Type I AGN (Shuntov et al., 7 Nov 2025). Morphology is defined through bulge-to-total light ratio measured in the rest-optical band closest to $\sim 0.2\,{\rm deg}^2$70–$\sim 0.2\,{\rm deg}^2$71m, with disk-dominated galaxies at $\sim 0.2\,{\rm deg}^2$72, intermediate systems at $\sim 0.2\,{\rm deg}^2$73, and bulge-dominated galaxies at $\sim 0.2\,{\rm deg}^2$74 (Shuntov et al., 7 Nov 2025).

The quiescent stellar mass function shows a rapid, mass-dependent build-up from $\sim 0.2\,{\rm deg}^2$75 to $\sim 0.2\,{\rm deg}^2$76, with the number density of the most massive quiescent galaxies increasing by $\sim 0.2\,{\rm deg}^2$77 dex at $\sim 0.2\,{\rm deg}^2$78 and then evolving little at $\sim 0.2\,{\rm deg}^2$79; by $\sim 0.2\,{\rm deg}^2$80 their density reaches $\sim 0.2\,{\rm deg}^2$81 (Shuntov et al., 7 Nov 2025). At low masses, a power-law upturn is detected out to $\sim 0.2\,{\rm deg}^2$82, which the paper interprets as indicative of environment-driven quenching (Shuntov et al., 7 Nov 2025). Both quiescent and star-forming populations are reported to have $\sim 0.2\,{\rm deg}^2$83 out to $\sim 0.2\,{\rm deg}^2$84, while bulge-dominated quiescent galaxies dominate the high-mass end of the quiescent stellar mass function at all redshifts and disk-dominated quiescent systems dominate only below $\sim 0.2\,{\rm deg}^2$85 (Shuntov et al., 7 Nov 2025).

Taken together, these results define the practical significance of COSMOS2025. It is publicly released with FITS tables, ASCII products, morphology tables, machine-learning class probabilities, photo-$\sim 0.2\,{\rm deg}^2$86 PDFs, LePHARE physical parameters, CIGALE non-parametric SFH vectors, and documentation and notebooks, and is explicitly presented as a ready-to-use resource for galaxy evolution and cosmological studies (Shuntov et al., 3 Jun 2025). The scientific record built on top of it already spans structural evolution, red-sequence formation, protocluster assembly, high-redshift dusty populations, and morphology-resolved stellar-mass functions, indicating that COSMOS2025 functions simultaneously as a survey catalog, a calibrated inference framework, and a reference dataset for JWST-era extragalactic studies.