An extremely lethal virus or other pathogen that acts on core cellular functions of all life.
Almost all living plants and animals have a core of basic genes that encode the basic cellular structure and biochemistry common to all life in Earth (i.e. they all share a last universal common ancestor).
For example, humans share 60% of their genes with bananas, fruit flies and chickens, respectively.
This last universal common ancestor and the shared ancestry that flows from it is one of the vulnerabilities of life on Earth to catastrophic failure.
If a lethal pathogen evolves naturally that implicates this shared ancestry and perhaps also had a long latency period to make its effects pass unnoticed until it is too late, every species could be put more or less equally at risk. The evolution of a pathogen fitting these criteria could have an impact as profound as the singular event in which life arose on Earth that spawned that last universal common ancestor, a once in a multi-billion year phenomena that nonetheless happened at some point purely through random chance.
This would be basically a Grey Goo scenario, modified to reflect that fact that mutations to resist it would eventually materialize.
A pathogen that targets the basic core of the genetic foundation of almost all life would affect almost all species of plants and animals on Earth and in the sea almost equally if a suitable transmission method were found.
For example, perhaps it would start by infecting Plankton, sea grass and algae in the sea, and grass and other foods at the base of the food pyramid on land. Then, the infection would be of a type that could be carried up the food chain.

A Marine Food Pyramid

A Terrestrial Food Pyramid
There are historical examples of naturally arising pathogens with some of these features although fortunately for us, not all of them. Bubonic plague and smallpox and HIV all had very high lethality before treatments were developed. Smallpox and HIV have affected multiple species. Mad cow disease naturally propagates up the food chain even in the face of normally sufficient sterilization methods. HIV has a long latency period. A perfect storm pathogen that randomly ends up combining the various features of these historic plagues could give rise to a mass extermination event if it operated by impairing the functioning of genes that are shared by almost all species.
It is relatively straight forward to tweak the theoretical lethality of a pathogen in a manner that would cause it to be thwarted by resistance mutations just in time for about 50% of any given set of species to be wiped out, although estimating the lethality in terms of numbers of species rather than numbers of members of a species is a bit of a leap and covers a domain of applicability of infectious agents that are much more lethal than any such agent now know on Earth.
Some of the foundations of this kind of analysis are set forth, for example, in:
Frank Hilker, et al., "Modelling virus coinfection to inform management of maize lethal necrosis in Kenya." Phytopathology (2017); DOI: 10.1094/PHYTO-03-17-0080-FI
Like most infectious agents that are highly lethal, sooner than later it would burn out, having killed all life that isn't immune to it.
Then, relatively quickly (a few centuries to a few million years - our scientific tools use to study previous events of this kind don't have the resolution needed to cabin the time range more exactly), the temporary oxygen shortage caused by the mass die off would be overcome and returned to stasis by the species that survived and expanded into the ecological niches of those that went extinct.