IBM published granite-speech-5.0-2b and revised versions of granite-speech-5.0-470m-turboctc and granite-speech-4.1-2b-plus, all Apache 2.0, with IBM-reported leaderboard and error-rate figures.
GitHub's October 26 changelog says assisted permissions, which use a language model to judge permission requests, are generally available; its docs say the feature is not a substitute for sandboxing.
InfoWorld and The Register report competing Postgres-plus-Iceberg designs from Databricks, EDB, Snowflake, pgEdge and AWS, with analysts flagging a shared DuckDB dependency and catalog gaps.
llm-d v1.0.0, published October 20, aligns component versions across its organization, moves the router configuration API to v1, adds DisaggregatedSet and deprecates its own ROCm image build.
Reporting on TypeScript 7, Vite 8, pnpm 12 and a Rust React Compiler shows native rewrites with large reported speedups, mostly from project benchmarks, and costs in size and compatibility.
Kotlin 2.5.0-Beta2, a pre-release published October 21, introduces experimental companion blocks and full value classes, a .kar publication format for multiplatform libraries, and an error on compiler runs below JDK 17.
Modular published the MAX graph compiler, runtime and hardware backends under a new MAX License that restricts AI-assisted reimplementation and distribution of modified source, while Apache 2.0 parts stay.
Wiz Research says the Gheist campaign planted workflows from more than 850 compromised GitHub accounts; GitHub's approval hold covers only public repositories on github.com.
GitHub's Copilot usage metrics API gained a totals_by_client array on October 23 and named custom skills, agents and MCP servers on October 22, with caveats on undercounting.
Helm published patch releases 4.3.1 and 3.22.1 on October 15. Both bump golang.org/x/net; the 3.22.1 changelog cites five Go vulnerability IDs, and the 4.3.1 notes cite none.
mise release v2026.10.19, published October 24, narrows what counts as project trust, enforces pinned checksums and SLSA provenance, and makes lockfile_mode generate the default, with a list of breaking changes.
OpenAI's API changelog says developers can now choose 12-hour prompt caching; its guide lists a 2.5x cache-write rate against 1.25x for the 30-minute default.