The day's clearest signal is model efficiency: Meta Superintelligence Labs released Muse Glimmer, a 30B-parameter Apache 2.0 open-weight model built for always-on local agents and coding, arriving alongside Nvidia's Nemotron 3.5 Lightning and NeMo Switchyard push into faster inference serving. Google and Modular both made language-level arguments for AI-assisted engineering, with Google pitching Go's simplicity for AI-generated code and Modular shipping Mojo 1.0. Security researchers flagged that OpenAI paused work on its unreleased Astra model after internal evals suggested it could cross a Critical cybersecurity capability threshold, a notable moment for frontier-model safety gating. On the infra side, ClickHouse shipped a major pg_clickhouse update claiming up to 1000x faster TPC-H queries via subquery pushdown, and OpenSSH cut a new point release.
- Meta ships Muse Glimmer, a 30B open-weight Apache 2.0 model for local agents
- OpenAI pauses Astra model work after Critical cybersecurity capability signals
- Nvidia launches Nemotron 3.5 Lightning and NeMo Switchyard for faster inference
- pg_clickhouse v0.10 claims up to 1000x faster TPC-H queries via subquery pushdown
- Modular ships Mojo 1.0; Google argues Go suits AI-assisted coding
Open-weight models and inference infra
Languages and developer tooling