TESS AI finds 118 new exoplanets

University of Warwick researchers said on Friday that an automated search of NASA’s TESS archive validated 118 exoplanets and produced a vetted list of more than 2,000 additional candidates. The results come from RAVEN, a pipeline that combines transit searches, machine-learning classification and statistical validation, according to a paper accepted for publication in *Monthly Notices of the Royal Astronomical Society*. (arxiv.org) The team said the search covered more than 2.2 million main-sequence stars observed by TESS in its first four years, across sectors 1 through 55. NASA has separately said TESS had discovered 679 confirmed exoplanets and 5,165 candidates by the end of September 2025, underscoring the size of the backlog that newer software is trying to sort through. ### Which data did the researchers search? The study said RAVEN worked on TESS full-frame image data from the mission’s first four years and focused on planets with orbital periods between 0.5 and 16 days. The target list covered more than 2.2 million main-sequence stars characterized with Gaia data, according to the paper’s abstract. NASA said in a separate January update that public TESS archives remain large enough for automated methods to keep finding likely planets. The agency said its own ExoMiner++ software identified 7,000 targets as exoplanet candidates in an initial TESS run, reflecting the scale of the screening problem. (arxiv.org) ### How did RAVEN decide which signals looked like planets? The paper said RAVEN first searched for transit-like dips with a box least squares algorithm, then used machine-learning models trained with realistic simulations to sort likely planets from false positives. (arxiv.org) The pipeline then applied statistical validation to produce several tiers of output, from newly validated planets to high-probability candidates still awaiting confirmation. Marina Lafarga and co-authors reported that process yielded 118 validated planets, including 31 that were newly detected in this search rather than recovered from earlier catalogs. (science.nasa.gov) The same run also produced about 1,000 new candidates inside the broader group of more than 2,000 high-probability cases. ### Why isn’t every candidate counted as a confirmed planet yet? NASA said an exoplanet candidate is a signal likely to be a planet but one that still requires follow-up observations with additional telescopes to confirm. That distinction matters because transit-like dips can also come from eclipsing binary stars and other astrophysical events, according to the agency’s description of its ExoMiner++ system. (arxiv.org) The RAVEN paper reflected that split by separating validated planets from candidates with high planet probability. The authors also flagged smaller groups of newly identified mono-transit and duo-transit candidates, along with large-radius candidates suited for further follow-up. (arxiv.org) ### How does this fit into the broader TESS search? NASA said on May 13 that TESS had mapped nearly 6,000 confirmed or candidate exoplanets across the sky as of September 2025. The agency said that tally included 679 confirmed planets and 5,165 candidates from the mission by the end of that month. (science.nasa.gov) The Warwick-led paper adds a uniform machine-led pass across a large TESS sample rather than a single-system discovery. The authors wrote that their vetted and validated samples were aimed at improving the candidate pool from TESS, especially for short-period planets. (arxiv.org) ### What happens next for the 2,000-candidate list? The paper said the unvalidated objects remain candidates and will need additional observing work to move into the confirmed-planet column. NASA has said such follow-up comes from other telescopes after the initial TESS detection. (science.nasa.gov) NASA also said ExoMiner++ has been released as open-source software on GitHub for researchers to use on TESS’s public archive. That means the next step in this search is likely to involve both new screening runs and telescope follow-up on the highest-priority RAVEN targets identified in the paper accepted by *MNRAS* in May 2026. (arxiv.org) (science.nasa.gov)