Community Pulse — From the Floor

A user reported on October 29 that Codex stopped a macOS desktop task after a context transition and ended by acknowledging standing instructions instead of continuing the work. The author closed the report as a duplicate to consolidate tracking under an existing issue, explicitly distinguishing that from a fix. We have not reproduced the behavior or established its prevalence.

In an October 29 diagnosis we have not independently reproduced, a user links a DeepSeek Harness Desktop 0.2.1-rc.3 startup failure on macOS to authentication cookies accumulating across restarts. The author reports an HTTP 431 response for oversized request headers, displayed by the app as an authentication failure, and says removing stale cookies restored startup. That recovery report does not establish that the vendor has fixed the product.

Mistral announced its Large 4 weight release on October 29, and its official repository lists weight files; we have not downloaded or run them. The Mistral Ecosystem Open-Weight License requires a separate commercial agreement for Model as a Service businesses’ commercial use, with exceptions including internal use and use through Mistral’s official products. A specialized cybersecurity variant offered to vetted partners in preview is a separate offering.

OpenAI’s October 29 Luna update aims for shorter answers and less agreeable language, but its system-card addendum reports higher factual-error rates than the October 7 version across all six reported metrics, while remaining below GPT-5.6 Luna. It also reports improvements in several under-18 safety categories and a decline in the biology dual-use refusal measure. These are company evaluations; they do not establish that concision caused the regressions.

Google announced expanded health collaborations and a research hub in Singapore on October 29. Its account distinguishes work with the National University Health System to evaluate an AI co-clinician from a population health program with NHG Health. The announcement describes collaborations and evaluation, without presenting an independent measurement of patient benefits or establishing completed nationwide clinical deployment.

Anthropic’s October 29 release notes introduce the Usage and Cost Analytics API in beta for Claude Console and Enterprise organizations, with reporting by dimensions including model, workspace and request user. Existing usage and cost reporting APIs are designated legacy but remain supported.