51 Eridani b isn't a new planet, but the software that found it in old Webb data could find new ones

Photographing a planet around another star means beating a blinding glare. An Earth-like world can be about 10 billion times fainter than the star it circles, Caltech says. Even a young giant planet is roughly 100,000 times fainter, according to Northwestern University.

A team from Northwestern, Caltech, MIT and the Space Telescope Science Institute (STScI) has found a new way through it. Their software, called FLIP, found the planet 51 Eridani b in older data from the James Webb Space Telescope, where the standard processing method had missed it. The study was published in Nature on 4 November 2026.

To be clear, 51 Eridani b is not a new planet. The Gemini Planet Imager, an instrument on a ground-based telescope, first photographed it in 2014, and it was announced in 2015. Webb itself imaged it in 2025. The news is the method, and what it might find next.

That matters because Webb’s archive holds years of data. If FLIP is as sensitive as the team reports, some of those images may still hide planets nobody has seen. So far, that is a goal.

How a digital twin pulled 51 Eridani b out of the glare

FLIP is short for Physics-based Learning for Imaging Planets. It builds a detailed computer copy, a “digital twin”, of Webb’s Near Infrared Camera (NIRCam). The software works out millions of tiny flaws in the telescope’s optics, predicts how the star’s light spreads across the image, and then removes it. What is left can be a planet.

FLIP borrows optimisation tools used to train AI models, Northwestern explains, but it is a physics model of one camera, not an AI that “spots” planets.

The team tested it on three stars: HIP 65426 and 51 Eridani, which have known planets, and TYC 5899, which has none. On the known planets, both universities report sensitivity 2 to 15 times better than standard techniques.

Excerpts from Caltech’s release, as published on Caltech’s newsroom on 4 November. Image: Caltech

For 51 Eridani b, Northwestern says the team used Webb NIRCam data about two years old, taken at a wavelength of 4.3 microns, where carbon dioxide absorbs light. The standard method, known as KLIP, did not detect the planet in that data. FLIP recovered it above 5 sigma, a level scientists treat as strong evidence that a signal is real.

The lead author is Rodrigo Ferrer-Chavez, a Northwestern PhD student, with Jason Wang (Northwestern) and Katherine Bouman (Caltech). NASA puts 51 Eridani about 97 light-years away, with its young gas giant orbiting roughly 890 million miles out, close to Saturn’s distance from the Sun.

Why even a steady telescope leaves starlight behind

Webb is stable to the nanometre, a millionth of a millimetre. Even so, tiny shifts leave starlight that standard subtraction cannot fully remove. That leftover glare is where faint planets hide.

Questions remain. Neither release explains exactly why the standard method missed, in this dataset, a planet Webb had already imaged in 2025. FLIP has not yet found a truly new world. And “2 to 15 times” is a wide range. In practice, a spread like that usually means the gain depends heavily on the target and the data.

What this changes about the James Webb planets you see

Webb’s pictures of planets are not simple snapshots. So how do astronomers find exoplanets this way? They take a real picture, called direct imaging, and then block or subtract the star’s light. Much of the work happens in software, after the telescope has done its part.

Some Webb results are not photos at all: its recent finding that the lava planet TOI-561 b reflects about 75% of its starlight came from patient light-counting, not a picture.

What to watch: whether the team’s archive search turns up anything new, and whether other groups test FLIP on their own data. The teams also want to use it on NASA’s Roman Space Telescope and on ground telescopes. Longer term, it could help plan NASA’s proposed Habitable Worlds Observatory, which needs about 100 to 1,000 times more stability than Webb, Northwestern says.

Better software could stretch Webb further than new hardware

The good news is real. FLIP gets more out of a telescope already in space, and testing it on a star with no known planets was a sensible check. Our reading: improving the software is far cheaper and faster than building a new observatory.

The limits matter too. This is one known planet, in one dataset. Some data details come from a single release. And headlines about a “James Webb hidden planet” oversell what happened.

On balance, this is a meaningful step in method, not a discovery.

The real test is Webb’s archive. If FLIP finds a planet nobody has seen before, and other teams confirm it, those headlines will finally be earned.