Author here. So we're currently at 4489 wins, 66 ties, and 197 losses vs simdjson + glaze across the "platform matrix" (macOS + GCC/CLANG, Ubuntu + GCC/CLANG, Windows + MSVC). In the following regimes:
Along with an exhaustive unit-test/correctness validation suite.
As well as position-for-position validation of our stage-1's output compared to simdjson across every test in the corpus.
Also be sure to check out the whitepapers we've posted, if you're interested. On the site under "Research".
Cheers!
I work with this guy, he probably knows more about C++ and JSON parsing than anyone I know, he built discord bots in C++ back in the day (before LLMs / vibecoding).
Author here. So we're currently at 4489 wins, 66 ties, and 197 losses vs simdjson + glaze across the "platform matrix" (macOS + GCC/CLANG, Ubuntu + GCC/CLANG, Windows + MSVC). In the following regimes:
Generic/Schemaless parsing (NEON, AVX2, AVX512)
Schema-aware parsing + serializing (NEON, AVX2, AVX512)
Stage-1 parsing (NEON, AVX2, AVX512).
Along with an exhaustive unit-test/correctness validation suite.
As well as position-for-position validation of our stage-1's output compared to simdjson across every test in the corpus. Also be sure to check out the whitepapers we've posted, if you're interested. On the site under "Research". Cheers!
I work with this guy, he probably knows more about C++ and JSON parsing than anyone I know, he built discord bots in C++ back in the day (before LLMs / vibecoding).