Benchmark is a topic tracked in our intelligence system with 5 linked articles.
Benchmark raises its first growth fund as part of a $2B capital raise, diverging from a long-standing fund-size tradition around $425M.
NVIDIA Cosmos 3 is an open-source physical AI foundation model family (Nano 16B and Super 64B) with a unified Reasoner-Generator MoT architecture, open datasets, post-training workflows, and production-ready NIM microservices for deployment.
Bijou64 is a canonical varint encoding using a two-trick design to ensure a single encoding per value, claiming 2–10x faster decoding than LEB128 with comparable wire size, and it’s released under MIT/Apache-2.0 with a CC BY-SA 4.0 spec, though it’s a nascent format not yet battle-tested.
European legaltech investors like General Catalyst, Y Combinator, Benchmark and Balderton are among the most active as funding in the sector accelerates, with year-to-date legaltech funding reportedly exceeding all of 2025.
A data-rich benchmark of Boost.Container’s segmented iterators shows substantial, compiler-dependent performance gains over non-segmented paths (up to ~17x in some cases), with mixed results from inner-loop unroll hints and stdlib implementations.
Microsandbox cut filesystem latency by 47× on average by moving from a host-led FUSE path to a prebuilt in-kernel EROFS disk image, delivering large benchmark gains and simplifying the codebase.
Subscribe for real-time topic updates and unlimited access to our intelligence platform.