On October 2 GitHub added confidential comments to repository security advisories. Only people with current write access can read them and get notified, which changes both the channel for talking to reporters and what REST-based integrations see.
Lemonade stable v2026.40.0 sizes streaming models on AMD integrated GPUs against the GTT pool. The same release dropped openmoss:rocm and sd-cpp:rocm on Windows, so upgrade problems need to be separated into the two.
llama.cpp and Ollama return typed decisions instead of generated text through /v1/systemone. Parsing gets simpler, but probabilities do not guarantee correct answers, and confidence cutoffs differ by model.
On September 30 NVIDIA announced general availability of the cuObject client and server libraries. Object payloads move to GPU or system memory over RDMA, but control requests stay on HTTPS/TCP, and storage servers must support the protocol too.
On October 1 Ai2 released Olmo-core 3, a framework with a redesigned mixture-of-experts (MoE) training system. Its trillion-scale tests show the system performance of the training stack, not the quality of a newly trained model.
PostgreSQL 19 Beta 4, released September 24, reverted several features originally planned for 19. Earlier beta feature lists no longer hold as is, and a major upgrade still needs pg_upgrade or dump and restore.
Python 3.10 reached end of life on October 1, and 3.10.22, released the same day, is its final security release. Applying 3.10.22 and moving to a supported Python are separate tasks.
Rust 1.99 stabilizes defining extern "C" and "C-unwind" variadic functions in Rust itself. Nightly is no longer needed, but matching the type and number of arguments the caller passes remains an unsafe contract.
Cloudflare added email one-time PIN protection to temporary Quick Tunnels. The --allowed-mail option is available in cloudflared 2026.9.3 and later, but non-browser clients and streaming remain constrained.
Claude Sonnet 5.5 keeps Sonnet 5’s base input and output rates, but changing the model ID alone can break a Messages API integration. Disabled thinking and forced tool_choice any/tool return HTTP 400; applications must also handle thinking blocks rather than assume the first response block is text.