Gaia Data Release 3: Summary of the content and survey properties

We present the third data release of the European Space Agency's Gaia mission, GDR3. The GDR3 catalogue is the outcome of the processing of raw data collected with the Gaia instruments during the first 34 months of the mission by the Gaia Data Processing and Analysis Consortium. The GDR3 catalogue contains the same source list, celestial positions, proper motions, parallaxes, and broad band photometry in the G, G$_{BP}$, and G$_{RP}$ pass-bands already present in the Early Third Data Release. GDR3 introduces an impressive wealth of new data products. More than 33 million objects in the ranges $G_{rvs}<14$ and $3100<T_{eff}<14500 $, have new determinations of their mean radial velocities based on data collected by Gaia. We provide G$_{rvs}$ magnitudes for most sources with radial velocities, and a line broadening parameter is listed for a subset of these. Mean Gaia spectra are made available to the community. The GDR3 catalogue includes about 1 million mean spectra from the radial velocity spectrometer, and about 220 million low-resolution blue and red prism photometer BPRP mean spectra. The results of the analysis of epoch photometry are provided for some 10 million sources across 24 variability types. GDR3 includes astrophysical parameters and source class probabilities for about 470 million and 1500 million sources, respectively, including stars, galaxies, and quasars. Orbital elements and trend parameters are provided for some $800\,000$ astrometric, spectroscopic and eclipsing binaries. More than $150\,000$ Solar System objects, including new discoveries, with preliminary orbital solutions and individual epoch observations are part of this release. Reflectance spectra derived from the epoch BPRP spectral data are published for about 60\,000 asteroids. Finally, an additional data set is provided, namely the Gaia Andromeda Photometric Survey (abridged)

third data release (Gaia DR3). The aim of this paper is to present the Gaia DR3 data products, providing a brief overview of the new features introduced in the processing and discussing the quality of the data. Gaia EDR3 (Gaia Collaboration, Brown et al. 2021) was the first instalment of the full Gaia DR3 and included astrometry and broad band photometry for a total of 1.8 billion objects based on 34 months of satellite operations. Radial velocities for 7 million sources were copied over from the second data release, Gaia DR2, where a small number of spurious radial velocities were removed (Seabroke et al. 2021).
Gaia DR3 complements Gaia EDR3, introducing a vast array of new data products based on the same source catalogue and raw observations at the basis of Gaia EDR3. Indeed, the astrometry and broad band photometry in G, G BP , and G RP from Gaia EDR3 are repeated in the new catalogue (but see section 3.1 for G correction). New data products in Gaia DR3 include mean low-resolution blue and red photometer (BP/RP) spectra and high-resolution Radial Velocity Spectrometer (RVS) spectra; new estimates of mean radial velocities to a fainter limiting magnitude; line broadening and chemical composition information derived from RVS spectra; variable-star classification and characterisation; and photometric time series for over 20 classes of variables. Photometric time series for all sources, variable and non-variable, are available for a field centred on the Andromeda galaxy. A large sample of Solar System objects (SSOs) with orbital solutions and epoch observations is part of the data release, together with reflectance spectra for a subset of those. Gaia DR3 also includes results for non-single stars (NSS), quasars, and galaxies. For a large fraction of the objects, the catalogue lists astrophysical parameters (APs) determined from parallaxes, broad band photometry, and the mean RVS or mean BP/RP spectra.
To enhance the scientific exploitation of the data, the Gaia archive includes pre-computed cross-matches with selected external optical and near-infrared photometric and spectroscopic surveys. In addition to the catalogues described in Gaia Collaboration, Brown et al. (2021), Gaia DR3 includes the sixth data release from the Radial Velocity Experiment (RAVE) survey (Steinmetz et al. 2020a). The Gaia Universe Model Snapshot (GUMS, version 20, Robin et al. 2012) and the corresponding simulated Gaia catalogue (GOG) were also already included in Gaia EDR3. The details on these supplementary data can be found in the online documentation 1 . This paper is organised as follows. Section 2 summarises the Gaia instruments and their wavelength ranges; section 3 summarises the properties of the astrometric and photometric data already included in Gaia EDR3; section 4 presents the BP/RP spectra; section 5 outlines the RVS data processing, listing the improvements implemented in Gaia DR3, and presents the new RVS data products, the line broadening velocity (v broad ) and the G RVS magnitude. Section 6 deals with the variable source content of the catalogue including the Gaia Andromeda Photometric Survey (GAPS); section 7 discusses the NSS content of the release; section 8 comments on the SSOs; section 9 provides information about the APs; section 10 presents the extended object (EO) data processing and results. Section 11 describes the extragalactic content of Gaia DR3, that is, information on galaxies and quasi-stellar objects (QSOs) or 'quasars' derived by several data processing modules. Section 12 comments on the quality of the release and section 13 provides references to the software 1 Gaia DR3 online documentation: https://www.cosmos.esa. int/web/gaia-users/archive/gdr3-documentation tools that are offered to the users to deal with Gaia DR3 data. Finally, section 14 presents some concluding remarks.
Each data product section briefly summarises the main limitations of the data, and makes reference to the relevant Gaia Collaboration and Gaia Data Processing and Analysis Consortium (DPAC) papers where more details can be found. For a number of technical details, we refer the reader to the online documentation.
All the papers accompanying Gaia DR3 are published in the Astronomy & Astrophysics special issue on Gaia DR3. Finally, we recall that all the Gaia data releases are made available through the archive hosted by ESA 2 , as described in Gaia Collaboration, Brown et al. (2021) and references therein. Partner and affiliated data centres in Europe, the United States, Japan, Australia, and South Africa provide access to the data through their own facilities.

Data processing
The Gaia satellite has three main instruments on board: the astrometric instrument collecting images in Gaia 's white-light Gband (330-1050 nm), the blue BP and red RP prism photometers for low-resolution spectra, and, finally, the RVS. BP/RP spectral data cover the wavelength ranges 330-680 nm and 640-1050 nm, respectively. The resolution is variable and ranges from 30 to 100 for BP and 70 to 100 for RP in λ/∆λ, depending on the position in the spectrum and on the CCD (see Carrasco et al. 2021). In the RVS spectra, the starlight is dispersed over about 1100 pixels in the Gaia telescopes scanning direction (along-scan, AL), sampling the wavelength range from 845 to 872 nm. The resolving power is R ∼ 11 500 (with a resolution element of about 3 pixels). As the wings of the spectra are excluded from the processing, the effective wavelength range of the processed spectra is reduced to 846-870 nm (Sartoretti et al. 2022a). Similarly to Gaia EDR3, Gaia DR3 is based on data collected over a 34-month time interval (for details see Gaia Collaboration, Brown et al. 2021). The Gaia data processing is the responsibility of DPAC and is presented in Gaia Collaboration, Prusti et al. (2016).
The basic statistics on the source numbers for each of the data products in Gaia DR3 can be found in Table 1, with further details in Table 2. The categories listed in the tables are described in the text below. A visual impression of the release contents is given in Fig. 1 which shows histograms of the distribution in G of the main categories of data products in Gaia DR3.

Astrometry and broad band photometry
The astrometric and broad band photometry content of Gaia DR3 is the same as that for Gaia EDR3, except for the addition of the G RVS photometry, but for convenience we summarise here the properties of these data products. Gaia EDR3 provided celestial positions and the apparent brightness in G for 1.8 billion sources. For 1.5 billion of those sources, parallaxes, proper motions, and the (G BP − G RP ) colour were also published. Gaia DR3 therefore contains some 585 million sources with five-parameter astrometry (two positions, the parallax, and two proper motion components), and about 882 million sources with six-parameter (6-p) astrometry, including an additional pseudocolour parameter. We refer to Gaia Collaboration, Brown et al. (2021) for details on Gaia EDR3, a summary of the astrometric A&A proofs: manuscript no. output  Table 2 Object DSC classification 1 590 760 469 Self-organised map of poorly classified sources 1 30 × 30 map and prototype spectra APs from mean BP/RP spectra 470 759 263 G < 19, see Table 2 APs from mean RVS spectra 5 591 594 Chemical abundances from mean RVS spectra 2 513 593 Up to 12 elements DIBs from mean RVS spectra 472 584 Non-single stars 813 687 astrometric, spectroscopic, eclipsing, orbits, trends, see  Eclipsing binaries 87 073 including eclipsingSpectro combined solutions and photometric data processing, and a summary of the limitations of these data and guidance on their use. Detailed descriptions of the astrometry in Gaia EDR3 (and thus also Gaia DR3) are provided in Lindegren et al. (2021b), while the broad band photometry is described in detail in Riello et al. (2021). Details Article number, page 5 of 23 GCRF-3 Solar system object transits The top panel shows the histograms for the basic observational data in Gaia DR3 (spectra, radial velocities, v broad , photometric time series). The middle panel shows histograms for the stellar astrophysical contents, and the bottom panel shows the non-stellar astrophysical contents. The sharp transitions in the top and middle panels at G = 17.65 and G = 19 are caused by the limit on the brightness of sources for which BP/RP spectra are published and the limit up to which astrophysical parameters were estimated. The SSO histogram shows the distribution of the transit-level G-band magnitudes (see Tanga et al. 2022, for a distribution of solar system objects in absolute magnitude H). The QSO and galaxy candidate histograms extend to very bright magnitudes which is a consequence of favouring completeness over purity in these samples, and not applying any filtering to remove them (see Gaia Collaboration,  on the construction of the Gaia DR3 source list can be found in Torra et al. (2021), while the basic inputs to the astrometric and G-band photometric processing (the G-band source image positions and fluxes) are described in Rowell et al. (2021). The validation of the astrometry and broad band photometry is summarised in Fabricius et al. (2021). The photometric pass-bands for G, G BP , and G RP are provided in Riello et al. (2021), and the one for G RVS in Sartoretti et al. (2022b).
Finally, the Gaia Science Alerts project and content are described in Hodgkin et al. (2021).
3.1. G-band photometry for 6-p and 2-p sources Section 7.2 of Gaia Collaboration, Brown et al. (2021) and Section 8.3 of Riello et al. (2021) describe a correction to be applied to the G-band photometry for sources with 6-p and 2-p astrometric solutions. This correction was provided in the form of Python code and Astronomical Data Query Language (ADQL) recipes. We note here that these corrections are included in Gaia DR3 and therefore should not be applied when working with photometry extracted from the Gaia DR3 data tables in the Gaia archive.
As noted in Gaia Collaboration, Brown et al. (2021) the Gband photometry for a small number of sources is not listed in Gaia EDR3. This issue has not been fixed for Gaia DR3. The magnitudes can be found in a separate table to be provided through the Gaia DR3 'known issues' web pages (for details see section 8.2 in Riello et al. 2021).

Celestial reference frame
Given that the astrometry in Gaia DR3 is unchanged from Gaia EDR3, it follows that the source positions and proper motions are provided with respect to the Gaia-CRF3, the third realisation of the Gaia celestial reference frame. Gaia-CRF3 is aligned with the third realisation of the International Celestial Reference Frame in the radio (ICRF3) to about 0.01 mas rootmean-square (RMS) at epoch J2016.0 (barycentric coordinate time, TCB), and globally non-rotating with respect to quasars to within 0.005 mas yr −1 RMS. The Gaia-CRF3 is defined by the positions and proper motions of 1 614 173 QSO-like sources that were selected using Gaia EDR3 astrometry. For the alignment and spin of the Gaia-CRF3, the special sets of 2007 and 428 034 sources, respectively, were used based on the preliminary astrometric solutions known as AGIS3.1 (Lindegren et al. 2021b). The construction and properties of the Gaia-CRF3, the comparison to the ICRF3, and the procedure to fix the alignment and spin of the astrometric solution are described in Gaia Collaboration, Klioner et al. (2022).

Systematic errors
The systematic errors present in the astrometry and broad band photometry published in Gaia EDR3 carry over to Gaia DR3. The conclusions from investigations during the data processing for Gaia EDR3 astrometry were that the global parallax bias for Gaia DR3, as measured from quasars, is −17 µas. The RMS angular (i.e. source to source) covariances of the parallaxes and proper motions on small scales are ∼ 26 µas and ∼ 33 µas yr −1 , respectively (see Lindegren et al. 2021b, for details). The parallax bias (and the proper motion systematic errors) varies as a function of magnitude, colour, and celestial position. This is extensively investigated in Lindegren et al. (2021a) and a recipe for correcting the parallaxes is given. The systematic errors in the broad band photometry are described in Riello et al. (2021).
Since the release of Gaia EDR3 several investigations of the systematic errors have been published. The parallax bias was investigated for specific sets of sources by Stassun & Torres (2021) The systematic errors in the proper motions and positions of stars in Gaia EDR3 were investigated by Cantat-Gaudin & Brandt (2021) and Lunz et al. (2021). Cantat-Gaudin & Brandt (2021) demonstrated that the proper motions of bright (G ≲ 13) stars show a residual spin with respect to the fainter stars by up to 80 µas yr −1 , and they provide a recipe to correct for this effect. Lunz et al. (2021) found differences in the alignment to the ICRF3 between bright and faint sources of about 0.5 mas. Systematic errors in the broad band photometry were investigated by Niu et al. (2021), Yang et al. (2021), and Thanjavur et al. (2021).
We stress here that including the above references does not imply an endorsement by the Gaia Collaboration of the results or any systematic error correction recipes provided in those papers.
Nevertheless, community efforts to investigate the quality of the Gaia data are highly appreciated and have in several cases led to an improvement in the validation procedures used during the Gaia data processing.

BP/RP spectra
BP/RP spectral observations are transmitted to the ground from the satellite in small windows surrounding the position of the source. The size of the BP/RP windows is 60 pixels AL by 12 pixels in the across scan (AC) direction, corresponding to an area in the sky of approximately 3.5 ′′ by 2.1 ′′ . The size of the window affects the detection of sources in crowded regions, resulting in partially overlapping windows. These windows are excluded from the Gaia DR3 data processing. A special treatment will be implemented in future data releases.

Data processing
The BP/RP processing is described in detail in De  and Montegriffo et al. (2022). The spectra are first calibrated to an internal reference instrument which is homogeneous across all devices, observing configurations, and time. This is achieved purely from BP/RP data for a sufficiently large subset of sources selected to cover all calibration units (windowing strategies, gates, magnitude ranges, time). The internal calibration removes a number of effects such as bias, background, geometry, differential dispersion, and variations in response and in the line spread function (LSF) across the focal plane (Carrasco et al. 2021;De Angeli et al. 2022). The internal reference system is linked to the absolute system (both in terms of flux and wavelength) via the external calibration, which is based on a dedicated catalogue of spectro-photometric calibrators (Pancino et al. 2021a). More details concerning the calibration of the BP/RP spectral data to the absolute reference system are presented in Montegriffo et al. (2022).
We release mean low-resolution BP/RP spectra for about 220 million sources. They are selected to have a reasonable number of observations (more than 15 CCD transits) and to be sufficiently bright to ensure good signal-to-noise ratio (S/N), that is, G < 17.65. To this list, we add a few samples of specific objects that could be as faint as G ∼ 21.43: about 500 sources used for the calibration of the BP/RP data, a catalogue of about 100 000 white dwarf candidates, 17 000 galaxies, about 100 000 QSOs, about 19 000 ultra-cool dwarfs, 900 objects that were considered to be representative for each of the 900 neurons of the self-organising maps (SOMs) used by the outlier analysis (OA) module (see section 9) and 19 solar analogues ).
The S/N of the BP/RP spectra varies depending on the magnitude and colour of the source. In the range 9 <G < 12, it can reach 1000 in the central part of the RP spectral range and it is of the order of 100 at G ∼ 15. These spectra have been extensively used as input for further data processing inside DPAC (see e.g. sections 8, 9, 12), which provides a strong validation of their exceptional quality.

Data representation
The source mean spectra are provided in a continuous representation: they are described by an array of coefficients to be applied to a set of basis functions. We use a set of 55 basis functions for BP and 55 basis functions for RP (referred to as 'bases') defined as a linear combination of Hermite functions (Carrasco et al. 2021;De Angeli et al. 2022). In low-S/N spectra, as for instance at faint magnitudes, it is possible that higher order bases are over-fitting the noise in the observed data. In particular, low-S/N spectra when sampled in pseudo-wavelength can exhibit unrealistic features (wiggles). To mitigate this problem, Carrasco et al. (2021) suggest a statistical criterion to select the coefficients that can be dropped without losing information (truncation). Non-truncated and truncated spectra are in agreement within the noise. It should be noted that sharp features in the spectra, such as emission lines, can only be reproduced using higher order bases and therefore imply a larger number of significant coefficients. As the rejection criterion is statistical, it might happen that too few or too many coefficients are removed. This might affect faint objects with sharp spectral features, such as QSOs, or emission line stars. The effects of the truncation for various object classes are described in Sect. 3.4.3 of De Angeli et al. (2022).
To allow the users to decide how many coefficients are relevant for their scientific case, all the 55 coefficients of the basis functions are released. The number of coefficients returned by the truncation criterion is given in the parameters bp_n_relevant_bases and rp_np_relevant_bases available in the xp_summary 3 table and in the mean continuous spectra available via Datalink 4 .
In addition to the continuous representation, sampled spectra representation (i.e. in the form of integrated flux vs. pixel) in both internal and absolute flux can be calculated using the Python package GaiaXPy made available with Gaia DR3 (see section 13). Sampling the spectra on a discrete grid in pseudowavelengths or absolute wavelengths results in a loss of information. In particular, the full covariance matrix is provided in the continuous representation, whereas it cannot be calculated in a spectrum sampled on a grid with more points than the number of basis function coefficients. Although sampled spectra are made available for a subset of sources in Gaia DR3, we strongly encourage users of the BP/RP spectra to make use of the continuous representation to maximise the scientific use of these data.

RVS data products
Gaia DR2 was the first release to include RVS radial velocities based on 22 months of data for stars at G RVS ≤ 12 and with effective temperatures 3500 < T eff < 6900 K.
Gaia DR3 contains newly determined radial velocities for about 33.8 million stars with G RVS ≤ 14 and with 3100 ≤ T eff ≤ 14500 K. Additional data products are published for the first time: v broad for about 3.5 million stars, G RVS magnitudes for more than 32 million stars, mean spectra for slightly less than 1 million stars, and epoch radial velocities for about 1000 RR Lyrae and roughly 800 Cepheids of different types ). All these data come with quality parameters. Users are advised to treat the Gaia DR3 radial velocity catalogue as completely independent of Gaia DR2.
The spectroscopic pipeline and the improvements since Gaia DR2 are described in Chapter 6 of the online documen-tation (Sartoretti et al. 2022a). All the products of the spectroscopic pipeline are available in the table gaia_source 5 , except for the mean spectra (see section 5.2). The transit radial velocities for the Cepheids and RR Lyrae stars are published in the table vari_epoch_radial_velocity 6 .
The products of the RVS pipeline, their properties, and their validation are described in more detail in dedicated papers: the radial velocity determination is described in Katz et al. (2022), the specific treatments to measure the hot star radial velocities in Blomme et al. (2022), and the radial velocity processing of the double-lined spectra is presented in Damerdji et al. (2022).
The v broad determination is discussed in Frémat et al. (2022) and the G RVS magnitudes and the RVS pass-band in Sartoretti et al. (2022b).
For each valid RVS spectrum entering the pipeline, the transit radial velocity is computed through a fit of the RVS spectrum relative to an appropriate synthetic template spectrum. In Gaia DR3, for stars cooler than 7000 K, the template input parameters are mostly taken from intermediate results of Apsis ) based on Gaia DR2 BP/RP spectra (see section 9). For hotter stars, they were derived as explained in Blomme et al. (2022).

Data processing improvements
A number of improvements were implemented in the Gaia DR3 RVS pipeline. Here we list the most significant. Bright stars are processed using the method already implemented in Gaia DR2, that is, the combined radial velocity for stars at G RVS < 12 is the median of the single-transit radial velocities. However, this method is not very efficient when the S/N is low. The radial velocities of faint stars in the range 12 < G RVS < 14 mag are obtained from the averaged single-transit cross-correlation functions referring to the Solar System barycentre.
To improve performance at the faint end, we implemented an improved stray-light-correction procedure. The correction map is estimated every 30 hours from the faint-star spectra edges and from the virtual objects 7 , while in Gaia DR2 a single stray-light map was computed from the Ecliptic Pole scanning law 8 data.
We introduce a deblending procedure when the transit spectra are contaminated by nearby sources falling inside the RVS window. These deblended spectra are then used to obtain radial velocities and mean spectra, while in Gaia DR2 they were simply removed from the pipeline. This allows us to process a larger number of epoch spectra per source, increasing the S/N. However, we point out that only the clean non-blended transits are used to derive G RVS , and v broad .
The LSF was calibrated in both the AL and AC directions. The LSF-AL calibration has reduced the systematic shifts between the two fields of view affecting the wavelength calibration zero point and the epoch radial velocities. The LSF-AC calibration was also used in the deblending procedure and in the es-timation of the flux lost outside the window for estimation of G RVS .
v broad is computed for each transit, excluding deblended spectra, and then averaged. In addition to the projected rotational velocity, v sin i, v broad can include other physical effects like macro-turbulence, residual instrumental effects (LSF model uncertainty), and template mismatches.
We consider that a spectrum can be contaminated by nearby sources with G RVS < 15 even if they are not located inside the RVS window. In that case, the target spectrum is removed when the contaminant differs from the target source by less than 3 mag. This effect was neglected in Gaia DR2.
Finally, G RVS (grvs_mag in gaia_source) is calculated as the median of the single-transit G RVS measurements. Values of G RVS fainter than 14.1 were regarded as spurious and removed because they could have been caused by an inaccurate background estimate.

RVS spectra
The published mean RVS spectra are identified using the column has_rvs in the gaia_source table. Their spectra are available through the Datalink interface in the table rvs_mean_spectrum 9 . The mean spectra and their processing are described in detail in Seabroke et al. (2022). In summary, the transit RVS spectra are extracted, cleaned, deblended (if needed), wavelength calibrated and normalised either to their pseudo-continuum or by scaling with a constant (the latter for cool stars or noisy spectra). The spectra are then shifted to the rest frame using the epoch radial velocities for the bright stars for G RVS ≤ 12 mag, or using the combined radial velocity for the faint stars; they are then interpolated into a common wavelength array spanning 846-870 nm with a step of 0.01 nm and averaged. S/N information is also provided in the table gaia_source.
The released RVS spectra are selected from among stars of spectral type AFGK with S/N > 20. In addition, a sample of low-S/N spectra spanning all spectral types is added (see Seabroke et al. 2022). A known issue is that the published spectra are not uniformly distributed over the sky.

Caveats
The contamination of the flux of an RVS target source by the flux of a nearby bright source can produce a spurious population of stars with high radial velocities. This issue was known from Gaia DR2 (Seabroke et al. 2021). In addition bright-star contamination produces biased radial velocities which do not necessarily stand out from the overall radial velocity distribution when two sources are separated by 1.8 ′′ in the direction perpendicular to the scan. These stars were removed from the catalogue (Katz et al. 2022).
Radial velocity uncertainties are generally very small, of the order of a few km s −1 or even less than 1 km s −1 at the bright end. However, they are slightly underestimated for bright stars. A correction is proposed in Babusiaux et al. (2022).

Variables
The approach to the variability analysis in successive data releases is iterative, including at each data release more variability types with lower signal to noise. In Gaia DR3 we publish a total of more than 10 million variable sources in about 24 variability types (and their time series), in addition to approximately 2.5 million galaxies. The two-dimensional structure of galaxies is observed by Gaia over a range of position angles. This can induce spurious (non-intrinsic) photometric variability. This effect is used to identify galaxies, but their time series are not released.
The Gaia DR3 variability content is a great leap in comparison with Gaia DR2 where we reached more than 550 000 stars, with six variability types.
In the variability pipeline (VARI) processing, the variability was first tested in the time domain, and was then characterised in the Fourier domain and classified by multiple classifiers. We refer to the documentation and to Eyer et al. (2022, and references therein) for more processing details. The input data are mostly the time series of field-of-view transits of the broad band photometry in the calibrated G, G BP , and G RP bands. Additionally, for Cepheid and RR Lyrae stars we use radial velocity time series from the RVS instrument (see section 5.2) Clementini et al. 2022); the longperiod variable (LPV) analysis included RP spectral time series (which are not part of Gaia DR3, Lebzelter et al. 2022); and short-timescale variables are based on per-CCD G-band photometry. We refer to Distefano et al. (2022) for more information about solar-like variables; Marton et al. (2022) for young stellar objects; Wyrzykowski et al. (2022) concerning microlensing events; Gomel et al. (2022) for ellipsoidal variables with possible compact object secondaries; and Carnerero et al. (2022) for information on active galactic nuclei (AGN) candidates. Candidate eclipsing binaries are described in Mowlavi et al. (2022). A subset of those also have parameters in the NSS tables (see section 7).
Variability products in Gaia DR3 can be accessed as follows. In the gaia_source table the field phot_variable_flag is set to 'VARIABLE' when a source appears in any of the vari_* tables, except for the vari_summary 10 table which includes non-variable sources that appear in GAPS (see below). The vari_summary table lists statistical parameters for all variable sources and all sources in GAPS. The time series in G, G BP , and G RP bands of all sources listed in the vari_summary table are available as light curve data through the Datalink interface.
The quality of this data base is impressive. For instance, concerning LPVs, the catalogue includes about 1.7 million sources with G variability amplitudes greater than 0.1 mag (0.2 mag in Gaia DR2). The period is derived for about 392 000 of them. In Gaia DR2 we only identified about 151 000 LPV sources. Concerning Gaia DR3 data, in many cases it was possible to identify the spectral class from RP spectra, that is, cool giants of spectral types M (oxygen-rich) and C (carbon-rich). This classification was based on the presence in the RP spectra of numerous molecular absorption bands, mainly those due to TiO and related oxides for M-type and molecular bands mostly associated with CH and C 2 molecules for C-type stars (Lebzelter et al. 2022).
More than 270 000 confirmed RR Lyrae stars are released in Gaia DR3, almost doubling the Gaia DR2 RR Lyrae catalogue. In addition, we provide a better characterisation of the RR Lyrae pulsational and astrophysical parameters. This, along with the improved astrometry published with Gaia EDR3, make this sample the largest, most homogeneous all-sky catalogue of RR Lyrae stars published so far .
A small, but significant number of micro-lensing event candidates (363 in total, of which 90 are new) are identified (Wyrzykowski et al. 2022). While testing the exoplanet detection method, two Gaia-discovered transiting extra-solar planets were found from the epoch photometry (Panahi et al. 2022). This demonstrates the feasibility of the detection approach and Gaia's potential for discovering exoplanet candidates. About 214 exoplanet candidates are released in Gaia DR3.
A large number of ellipsoidal variable candidates were detected. Their variability is due to the tidal interaction with a companion in a close binary system. About 6 000 short-period ellipsoidal variables have relatively large-amplitude modulations in G, possibly indicating a massive, unseen secondary. Among those, 262 systems have a higher probability of having a compact secondary. Follow-up observations are needed to verify the true nature of these variables (Gomel et al. 2022).

Gaia Andromeda Photometric Survey
The photometric time series for all Gaia sources located within a 5.5 • radius centred on the Andromeda galaxy are part of GAPS, which contains more than 1.2 million sources. Whether or not a source appears in the GAPS survey is indicated in the gaia_source table by setting the field in_andromeda_survey to 'true'. The time-series statistics for GAPS sources are available in the vari_summary table. Evans et al. (2022) give more details.

Caveats
As a caveat, a number of sources show more than one type of variability. In general this overlap can be scientifically explained. However, this is not always the case, for instance a few objects were classified as both long period and short timescale (Lebzelter et al. 2022). Holl et al. (2022) discuss spurious periodic variations in the photometric data, due to instrumental effects.

Non-single stars
About 800 000 solutions for NSS including astrometric (Halbwachs et al. 2022), spectroscopic (single lined SB1; and double lined SB2) , and eclipsing binaries (Siopis et al. 2022) are published in Gaia DR3, with either orbital elements or trend parameters, or combinations of these.

NSS archive tables
The NSS tables are organised according to the type of solution: nss_two_body_orbit 11 contains the orbital parameters for all the binary categories; nss_acceleration_astro 12 contains accelerations for sources with an astrometric motion better described using a quadratic or cubic proper motion; nss_non_linear_spectro 13 presents trend (long period) solutions of spectroscopic binaries; nss_vim_fl 14 includes objects that exhibit photocentre displacements due to the photometric variability of one component, requiring the correction of the astrometric parameters.
This catalogue outnumbers existing surveys by large factors, spanning a large range of binary types, periods, and magnitudes. The potential of the Gaia DR3 binary star content is outlined in Gaia Collaboration, .

Caveats
Binaries can simultaneously belong to different classes; for example, astrometric binaries can also be spectroscopic binaries (identified as astroSpectroSB1 in the nss_solution_type in the table nss_two_body_orbit) or eclipsing binaries can also be spectroscopic binaries (identified as eclipsingSpectroSB1). In many cases of multiple solutions, combined solutions have been computed and included in the Gaia DR3 catalogue. However, combined solutions are not always provided, and sources can be found in several tables simultaneously. More information and advice on how to deal with these cases can be found in Gaia Collaboration, .
Acceleration solutions are not always in agreement with expectations from known orbits in external catalogues, and a fraction of them could have had an orbital solution. In general the parallaxes and proper motions derived by the NSS processing are more precise than those derived by the Gaia astrometric global iterative solution (see section 3) which assumes that all stars are single. This is not always the case for the acceleration solutions, which should be used with caution (Gaia Collaboration, Arenou et al. 2022;Babusiaux et al. 2022).
Spurious solutions around the satellite precession period (62.97 days) or for some short periods can be found for SB1. Formal uncertainties are not rescaled according to the goodness of fit for all the binary types, but only for the astrometric solutions. The NSS sample is far from complete. This is because of a number of selection effects due to data processing and additional filtering. Statistical studies on the data should take this into account.

Solar System objects
In Gaia DR3, about 160 000 SSOs were processed and analysed. As in previous data releases, known SSOs are searched for by matching the observed transits to computed transits based on the information on the satellite orbit, the scanning law, and a numerical integration of the SSO motion. Tanga et al. (2022) gives a description of the selection and processing. After filtering the list for possible contaminants, the final input selection had 3 513 248 transits for 156 837 known asteroids. Planetary satellites were also added following a similar procedure. In total, 31 planetary satellites are included. In addition, Gaia DR3 includes the astrometry of unknown moving sources based on the AL motion of objects observed from December 2016 to June 2017. The final input list of unidentified SSOs for Gaia DR3 comprises 4522 transits, corresponding to 1531 groups of chained transits of objects that at the time of processing were considered unmatched. Later, Tanga et al. (2022) identify 712 SSOs. These sources still appear as unmatched in the table sso_source 15 . It cannot be excluded that some of the still unmatched sources can be linked to known objects.
The astrometric accuracy of the orbits is impressive and remains at sub-mas level for G < 17, reaching an exceptional value of ∼ 0.25 mas for 12 <G < 15 mag.
Gaia DR3 contains spectro-photometry for more than 60 000 asteroids the majority of which have been observed with G between ∼ 18 and 20. The internally calibrated BP/RP epoch spectra are divided by the solar analogue spectrum to obtain epoch reflectance spectra that are subsequently averaged and sampled in 16 bands in wavelength. Only spectra derived using more than three epochs and with an average S/N of higher than 13 are published. No further rejection is applied. Poor spectra are flagged on a wavelength-by-wavelength basis introducing the sso_reflectance_spectrum_flag 16 , an array of 16 integers, one for each wavelength of the spectral bands (Gaia Collaboration, . The main properties of the reflectance spectra are described in section 12.

Object characterisation
Gaia DR3 includes APs for stars, galaxies, and QSOs. About 1600 million objects have class probabilities (G < 21), about 470 million stars have stellar parameters (G < 19), and there are about 6 million QSOs and about 1.3 million galaxy candidate redshifts. More details can be found in Creevey et al. (2022) concerning the AP content and an overview of the methods used in the software (Apsis) to produce these data.

Data processing
In total, 13 modules in the Apsis software provide 43 primary APs along with auxiliary parameters in 538 fields which appear in ten tables of the Gaia archive. A subset of APs is available in gaia_source. The astrophysical characterisation makes use of Gaia EDR3 broad band photometry and parallaxes, and Gaia DR3 mean RVS spectra and internally calibrated sampled mean BP/RP spectra. The stellar APs comprise atmospheric properties, evolutionary parameters, metallicity, individual chemical element abundances, and extinction parameters, along with other characterisation such as equivalent widths of the Hα line and activity index for cool active stars.
The discrete source classifier (DSC) produces the object classification, that is, it assigns class probabilities to all sources for five main classes, using different classifiers; these classes are QSO, galaxy, star, white dwarf, and physical binary star. The result of this classification is then used by four modules to initiate the processing of galaxies, QSOs, outliers, and extinction. The unresolved galaxy classifier (UGC) and the QSO classifier (QSOC) modules ) provide redshifts for candidate galaxies and QSOs. A more detailed discussion about the Gaia DR3 extragalactic content can be found in section 11.
The OA module performs an unsupervised classification for sources with lower probabilities from DSC, using SOMs (Kohonen 2001). OA groups similar objects into neurons on a 30 × 30 grid according to the similarity of their BP/RP spectra, as reported in oa_neuron_information 17 (see Creevey et al. 2022, figure 11, for examples). Other multi-dimensional data comprise a total Galactic extinction map at four healpix levels as well an optimal-level map derived by the total Galactic extinction module (TGE). The above data products are detailed in Delchambre et al. (2022).
For the stellar and interstellar medium characterisation, the General Stellar Parametriser for Photometry (GSP-Phot) derives astrophysical parameters (T eff , log g, A G , distance, . . . ) from BP/RP spectra down to G = 19 assuming they are all stars (Andrae et al. 2022). The multiple source classifier (MSC) derives similar stellar parameters under the hypothesis that all objects are unresolved binaries using the same input data as GSP-Phot. The General Stellar Parametriser for Spectroscopy (GSP-Spec) derives atmospheric parameters using the mean RVS spectra. It additionally derives 13 chemical species and diffuse interstellar band (DIB) equivalent widths (see Recio-Blanco et al. 2022, for details). Other modules analyse only a selected class of objects. This is the case for the Extended Stellar Parametrisers (ESPs) dealing with emission line stars (ESP-ELS), hot stars (7000 < T eff < 50 000 K. ESP-HS), cool stars (ESP-CS), and ultra-cool dwarfs (T eff < 2500 K, ESP-UCD). These modules together produce T eff and log g but also v sin i, spectral type classifications, Hα equivalent widths, and chromospheric activity index (using the calcium infrared triplet in the RVS domain (Lanzafame et al. 2022).
The Final Luminosity Age Mass Estimator (FLAME) derives evolutionary parameters from data that are processed by GSP-Phot and GSP-Spec. These comprise the stellar radius, luminosity, gravitational redshift, mass, age and evolutionary stage for stars. All of these data products appear in the astrophys-ical_parameters 18 and astrophysical_parameters_supp 19 tables in the Gaia archive. In addition to these data products, Markov Chain Monte Carlo (MCMC) samples are provided for two of the modules, GSP-Phot and MSC through the Datalink interface.
To date, this is the most extensive catalogue of astrophysical parameters homogeneously derived. It is based on Gaia-only data, and it will be superseded only by the fourth data release, Gaia DR4. Until then, it will remain the reference for upcoming ground-and space-based surveys. References to some applications and advice on the use of the APs can be found in the papers listed in section 12.

Caveats
The AP quality, although quite reliable on average, does vary depending on the quality of the input data (parallaxes, magnitudes, and spectra) but it can also vary based on the assumptions in the methods. The users should be aware of several issues. Variability is not taken into account in the processing, because only mean spectra and mean broad band magnitudes are used as input. Therefore, parameters of stars showing large variability might be inaccurate.
A&A proofs: manuscript no. output APs derived for objects in high-density regions are affected by crowding. For instance [M/H] in the core of a dense cluster can significantly differ from the value in the outskirts.
GSP-Phot distance estimates are underestimated for sources with large parallax errors because of an extinction prior which then dominates the distance inference. GSP-Phot metallicity is poorly derived for metal-poor stars ([M/H] < −1), and a metallicity calibration tool is proposed to the users to calibrate these. The data show residual degeneracy among the parameters, for instance T eff and the extinction in the G, A G , from GSP-Phot.
MSC treats all stars as unresolved binaries in BP/RP spectra; DSC shows poor performance on the classification of physical binaries and white dwarfs.
Additional information about the limitations of the APs are detailed in Babusiaux et al. (2022), while both the quality and validation of the stellar and non-stellar content are discussed more extensively in Fouesneau et al. (2022) and Delchambre et al. (2022), respectively. In general, users are advised to use quality flags and follow the recommendations detailed in the above papers.

Extended objects
The Gaia on-board system is designed to detect point-like sources (de Bruijne et al. 2015). However, EOs such as galaxies and galaxies hosting a QSO can be detected if their central region is sufficiently compact and bright. In this case, it is possible to reconstruct a two-dimensional light profile of the extended source. At each passage Gaia observes the nine one-dimensional Astro Field (AF) windows and the two-dimensional Sky Mapper (SM) windows. As the scan angle changes from one observation to the next, after a sufficient number of transits a large part of the source is covered by the observations. Dedicated software can combine the observed windows on ground to infer the twodimensional light profile parameters. During the first 34 months of operations about 116 million transits in the AF and SM focal planes were collected and processed. The EO pipeline compares the observations with a large number of simulations of galaxy images deriving the best-fit surface brightness profile parameters.
The lists of objects that were processed are derived from external catalogues. The list of QSOs was set up by merging the major catalogues of QSOs or AGN candidates published before 2018. In addition, we make use of an unpublished selection of candidates based on Gaia DR2 QSOs, and classified on the basis of their photometric variability (Rimoldini et al. 2019). The list of galaxies was derived from a preliminary analysis of Gaia DR2 sources with a match in the allWISE catalogue (Cutri et al. 2013). An unsupervised heuristic method (Krone-Martins et al. 2022) selects sources for a subsequent analysis. More than 1 million previously identified QSOs are analysed, identifying a host galaxy around approximately 64 000 of them. The surface brightness profiles of the host is published for a subset of about 15 000 QSOs, with a robust solution. Their Sérsic indexes indicate that they are mostly disc-like galaxies. Concerning the galaxy sample, two profiles are published (a free Sérsic profile and a de Vaucouleurs one). About 940 000 galaxies were analysed and robust solutions were derived for about 914 000 of them. The distribution of the parameters indicates that Gaia mostly detects elliptical galaxies.
These data can be found in the Gaia DR3 tables qso_candidates 20 and galaxy_candidates 21 . Details on the data processing and input lists can be found in Ducourant et al. (2022). This impressive data base is the first all-sky catalogue of two-dimensional light-profile parameters of galaxies, and QSO host galaxies derived at a resolution of about 200 mas.
The use of input lists clearly favours purity over completeness, even if a residual contamination by stars is still present.
Gaia observations are limited to the central area of the galaxy. For this reason, the derived brightness profile parameters do not always agree with external catalogues Ducourant et al. 2022).

Gaia DR3 extragalactic content
Extragalactic objects are classified or analysed by several modules in the Gaia data-processing system using various input data and methods. An overview of the extragalactic processing and content in Gaia DR3 is given in Gaia Collaboration, .
A predefined list of objects is analysed to derive the surface brightness profiles by the EO module. Classification of extragalactic objects is performed independently by two modules: the VARI module uses photometric light curves, whereas DSC uses the BP/RP spectra and astrometry. The modules QSOC and UGC estimate the redshifts of QSO and galaxy candidates (respectively) identified by DSC. Finally, the OA module performs a clustering of low-probability classifications from DSC.
The fact that different methods, input data, and training data are used to classify or select extragalactic objects has an important consequence: there is no common definition of QSOs or galaxies across the various Gaia modules. DSC and OA focus on completeness rather than purity. UGC processes objects with a higher probability of being galaxies according to the DSC classification, whereas QSOC uses a very low threshold on the DSC QSO probability, in order to analyse as many sources as possible. As a result, the sample of QSO candidates with redshifts is complete rather than pure, whereas that of the galaxy candidates with redshifts is of higher purity, although some contamination remains (see below).

Extragalactic object tables
Each one of the above modules provides independent results that are available across several tables. The table astrophysi-cal_parameters lists all the parameters produced by DSC. The oa_neuron_information contains the SOMs from the OA module. The tables vari_classifier_result 22 and vari_agn 23 present the parameters of the AGNs identified through the photometric light-curve analysis. From the above tables, we produce two integrated tables in Gaia DR3, qso_candidates and galaxy_candidates, where all good QSO and galaxy candidates are listed. The selection rules are detailed in Gaia Collaboration, . The gaia_source table includes the DSC classification probability as well as two flags, in_qso_candidates and in_galaxy_candidates, which indicate the presence of the source in the integrated table(s).
Already in Gaia EDR3, a list of 1.6 million compact extragalactic sources was used to define the Gaia-CRF3 (Gaia Collaboration, Klioner et al. 2022). These sources were identified from Gaia EDR3 astrometry by means of positional cross-matching with 17 external catalogues of QSOs and AGNs, and were subsequently filtered using astrometric criteria to remove stellar contaminants. The sample spans a magnitude range 13.4 < G < 21.4 with a median positional error of 0.45 mas. By construction this sample is of high purity.

Completeness and purity
The integrated tables of extragalactic objects favour completeness over purity. qso_candidates contains 6.6 million candidate QSOs and galaxy_candidates contains 4.8 million candidate galaxies, with global purities estimated to be 50% and 70% respectively. Contamination is mostly due to fact that there are many more stars in the Gaia survey, and arises in part from the choice of probability classification threshold for inclusion. Furthermore, some modules chose not to filter out regions such as the Magellanic Clouds and the Galactic plane, which are dense regions of stars and therefore also of contaminants. A very small fraction of sources are too bright to be genuine QSOs or galaxies. The same astrometric filtering as for Gaia-CRF3 applied to the content of the integrated QSO table qso_candidates gives a set of ∼ 1.9 million sources, which are identified by the flag astro-metric_selection (Gaia Collaboration, Bailer-Jones et al. 2022). Purer subsets of the two tables, containing ∼ 1.9 million probable QSOs and ∼ 2.9 million probable galaxies (both with around 95% purity) can be selected using the queries on EO, DSC, and VARI quality flag parameters listed in Gaia Collaboration, . Concerning QSOs, this new sample has about 1.8 million sources in common with the astrometric sample.
Finally, the DSC star class could be used to select a purer star sample, rejecting objects that are in the qso_candidates or galaxy_candidates, that is, those identified as candidate galaxies, QSOs, or extended objects. However, concerning extended objects, the DSC has not used morphological information in its classification (see for details Delchambre et al. 2022). Indeed there is no general morphological classification in Gaia DR3. Only a limited number of galaxy and QSO candidates (based on non-Gaia data) were analysed for evidence of a host galaxy using 2D morphology built up from the 1D scans (see section 10). As a consequence, residual contamination from extragalactic sources can still be present in a star sample selected following the above criteria.

The scientific performances of the catalogue
The scientific quality and the potential of the Gaia DR3 data are demonstrated in a number of accompanying papers providing basic science applications and additional validation of the catalogue. These illustrate the limitations of the data, providing advice and guidance to the users on specific scientific problems. The following topics are treated.
Gaia Collaboration, Creevey et al. (2022) present clean samples of very high-quality astrophysical parameters of stars derived from low-resolution BP/RP and RVS spectra all across the HR diagram, selected through severe quality cuts. These data have a long-term legacy value for future follow-up studies and missions and are released together with Gaia DR3 in separate tables. These samples include about 3 million OBA young disc stars, roughly 3 million FGKM stars, and about 21 000 ultracool dwarfs. Gaia Collaboration, Creevey et al. (2022) identify specific subsamples of particular interest to the community such as approximately 6 000 solar-analogues and 15 000 carbon stars, and provide homogeneous parameters for a subsample of the 111 Gaia spectro-photometric standard stars defined by Pancino et al. (2021b). For the ultra-cool dwarfs Gaia and 2MASS data are combined to provide radius and luminosity.
Gaia Collaboration, Recio-Blanco et al. (2022) use Gaia astrometry, radial velocities, chemo-dynamical analysis of disc and halo populations, producing the largest all-sky chemical map to date and analysing the abundances of streams from accretion events. This map includes chemical information for 597 open clusters, which represents the largest compilation of abundances homogeneously derived for clusters.
Gaia Collaboration, Schultheis et al. (2022) discuss the distribution of the DIB at 862 nm in the RVS spectral range in connection with the interstellar extinction to within a few kiloparsecs of the Sun. The paper provides the most precise measurement to date of the rest frame wavelength of the DIB at 8620.86 ± 0.019 Å.
Gaia Collaboration, Drimmel et al. (2022) explore nonaxisymmetric features in the disc of the Milky Way in both configuration and velocity space. These authors use Gaia DR3 APs and variability classifications to select various stellar populations in the disc of the Milky Way, tracing the spiral structure up to 4-5 kpc from the Sun. About 6 million red giant branch stars allow the authors to map the velocity field in the disc up to 8 kpc from the Sun, allowing the detection of signatures of the Galactic bar.
Gaia Collaboration, Arenou et al. (2022) present a clean catalogue of binary stars, discussing its completeness and some statistical features of the orbital elements in comparison with external catalogues. In addition, a catalogue of tens of thousands of stellar masses is provided. Several compact object candidates are identified. The catalogue includes sources found in rare evolutionary stages such as EL CVn, underlining the potential of analysing both photometric and orbital data. New binary UCD candidates are discovered and their masses estimated; two new exo-planets are found, and several dozen candidates are identified. The catalogue is made available together with Gaia DR3 tables.
Gaia Collaboration,  present reflectance spectra, derived from BP/RP spectra, for SSOs, discussing their statistical properties. The (z − i) colours are derived from the spectra, where z and i represent the values of the reflectances estimated with spline interpolation at 748 and 893 nm, respectively. (z − i) allows the identification of asteroid families sharing similar properties and common origin. An interesting feature is that the spectral slope 24 of the mean reflectance spectra and the depth of the 1 µm absorption band seem to show an increase with time for the S-type family, which is possibly due to space weathering effects on the asteroid surfaces. The catalogue is included in the data release.
Gaia Collaboration, Montegriffo et al. (2022) demonstrate the potential of synthetic photometry obtained from fluxcalibrated BP/RP spectra for pass-bands fully enclosed in the 24 the spectral slope is derived by fitting the mean reflectance spectra in the wavelength range 450 and 760 nm with a straight line and taking the angular coefficient Gaia wavelength range. This photometry has been used internally for validation purposes. Synthetic photometry in various photometric systems can be produced using the Python package GaiaXPy (see section 13). High-quality external photometry in large and medium passbands is reproduced at a few percent level in general and up to the milli-mag level when the synthetic photometry is standardised using an external reference catalogue. For a subset of 13 wide and medium bands, we release the Gaia Synthetic Photometry Catalogue (GSPC), an all-sky space-based catalogue of standardised photometry for the majority of the stars with released spectra and G < 17.65. A separate catalogue contains synthetic photometry in a selection of relevant bands for 100 000 white dwarfs selected from Gaia EDR3. Lastly, Gaia Collaboration, De Ridder et al. (2022) investigate the properties of high-mass main sequence pulsators, showing that Gaia DR3 data are precise and accurate enough to identify nearby OBAF-type pulsators.

Software tools
A number of Python software tools 25 are made available to the scientific community. The following are some examples.
-The Python package GaiaXPy which offers the possibility to compute synthetic photometry from BP/RP spectra in various photometric systems (see Gaia Collaboration, Montegriffo et al. 2022, and Sect. 12), and to simulate BP/RP mean spectra from a custom input absolute spectral energy distribution. In addition sampled mean spectra can be derived from the continuous basis function representation. Detailed information can be found in De . -A tool to re-calibrate the GSP-Phot metallicity output (see Creevey et al. 2022, Appendix E for detail) based on comparison with external surveys. We point out that the GSP-Phot and GSP-Spec re-calibration tools proposed here have their own limitations, as described in the related documentation. They provide tentative expressions to be used at the discretion of the user, depending on the science case.

Conclusions
Gaia DR3 expands Gaia EDR3 with a rich and large set of data products containing detailed astrophysical information for the same source catalogue. The quantity, quality and variety of astrophysical data constitute a major advance in the series of Gaia data releases. This release includes the largest collection of allsky spectrophotometry and radial velocities and the largest collection of variable sources ever produced. The availability of APs from low-and high-resolution spectra greatly surpasses existing catalogues. This catalogue includes a spectrophotometric and dynamical survey of SSOs of the highest accuracy. The NSS content outnumbers all existing binary star catalogues, and Gaia DR3 also contains the first all-sky space-based survey of QSO host and galaxy two-dimensional light profiles.
In this paper we briefly summarise the main additions and improvements to the data processing and we include comments on the data quality. However, the Gaia DR3 data products are numerous and complex and not all the details could be presented here. A more complete overview, insights into known issues with the data, and advice on the use of the data can be found in the papers accompanying the release and in the online documentation.
The nominal Gaia mission ended on July 2019. The mission has been extended since then to the end of 2022, with an indicative approval to extend the mission to the end of 2025. In 2025, the propellant for the micro-propulsion system is expected to be exhausted, and the precision on the attitude and spin rate of the satellite required for the astrometry can then no longer be maintained. At this point, the Gaia mission end of life will be reached and over 10 years of data will have been collected. In this context, two further data releases are foreseen, Gaia DR4 and Gaia DR5 which will include the data from the extended mission. Gaia DR4 will be based on 66 months of data, including a six-month period when the satellite was operated with a reversed direction of the precession of the spin axis around the direction to the Sun. This will mitigate the degeneracy between the across scan motion of the sources and their parallaxes, reducing this specific source of systematic errors on the astrometry. A major new aspect of Gaia DR4 is that it will include all time series data, that is, epoch astrometry, broad band photometry, radial velocities, and epoch BP, RP, and RVS spectra for all sources. In addition, we plan to release full astrometric, photometric, and radial-velocity catalogues; all the available variablestar and NSS solutions; an extended sample of source classifications and multiple APs for stars, and extragalactic objects. Finally, an updated extra-solar planet list will be published. The publication of Gaia DR4 is expected not before the end of 2025. Gaia DR5 will be the final release from the Gaia mission, and will be based on data collected over the full nominal plus extended mission periods and including all the data products mentioned above, as well as the Gaia legacy archive. Gaia DR5 is expected not before the end of 2030. Figure 2 summarises the astrometric uncertainties of the Gaia data releases so far and also shows the extrapolated uncertainties for Gaia DR4 and Gaia DR5. The latter are extrapolated from the Gaia DR3 performance according to the amount of data collected, 5.5 years and an expected 10 years, respectively 26 . With time the signal-to-noise ratio for all Gaia data products, including parallaxes, improves as √ t, and therefore the precisions for Gaia DR4 and Gaia DR5 are expected to improve by factors of 1.4 and 1.9, respectively. For the proper motions, the improvement goes as t 1.5 , which means improvements by factors of 2.7 and 6.6 with respect to Gaia DR3. For a more extensive discussion of the expected gains in future Gaia data releases, we refer to .
There is therefore still much more to look forward to, but for now we invite the reader to explore the veritable supermarket of astronomical and astrophysical information that is Gaia DR3.

Appendix A:
The Gaia mission and data processing have financially been supported by, in alphabetical order by country: -