Pyston was started at Dropbox in 2014 by Kevin Modzelewski and colleagues to reduce the costs of Dropbox's rapidly growing Python server fleet. Pyston v1 was a from-scratch implementation of Python 2.7 featuring a conservative tracing garbage collector and LLVM-based compilation tiers. By version 0.6.1, it achieved nearly 2x CPython performance on some benchmarks and 48% faster on web workloads, but compatibility with existing Python packages remained a major challenge.

In 2017, Dropbox ended its involvement with Pyston, choosing instead to rewrite performance-critical code in Go and other languages. The project appeared dead. However, in 2019, lead developers Kevin Modzelewski and Marius Wachtler regrouped independently and took a radically different approach: rather than reimplementing Python from scratch, they forked CPython 3.8 directly. This became Pyston v2, released in late 2020, achieving approximately 20% faster performance than stock CPython 3.8.

Subsequent releases improved performance significantly: Pyston v2.2 claimed 30% faster on web server benchmarks, and later versions reached up to 66% total speedup over stock CPython. The project also offered 'Pyston Lite,' a lighter-weight extension achieving roughly 10% speedup. In August 2021, the Pyston team joined Anaconda, which provided funding and packaging expertise.

The Pyston GitHub repository is now marked as 'no longer maintained,' with the cinderx project from Meta being referenced as a related effort. Pyston's journey mirrors a broader pattern: corporate-funded Python performance work that struggles to sustain itself independently.

Demonstrated that meaningful speedups are achievable by forking CPython and adding JIT compilation. Contributed to the broader conversation about Python performance that led to the official Faster CPython project. The v2 approach of forking CPython rather than reimplementing from scratch proved more practical than the v1 from-scratch approach.