$ the-wire · showcase
gpt-6-rosalind enters SDK model types as Codex hardens Cloud threads
By RepoJournal · Filed · About OpenAI · Composed from the cited sources · methodology
2 people shipped this
openai-node and openai-python both added gpt-6-rosalind to their beta Responses model values, while a cluster of Codex TUI fixes targets duplicated messages, stale disconnects, and missing transcripts when resuming Cloud threads.
Treat goal objectives as untrusted context openai/codex
Goal objectives now travel as untrusted goal_objective context rather than being embedded in continuation instructions, and the model is told to call get_goal when the objective is missing or truncated. Newly created objectives get injected into their originating active turn without dropping unrelated client context, and render as tool outputs when UntrustedAsToolOutput is enabled.
feat(api): recognize gpt-6-rosalind in beta Responses (#3126) openai/openai-node
Both SDKs now accept gpt-6-rosalind among known beta Responses model values, in openai-node and in openai-python's type hints. Typed clients pinning to a known-models union need the release that includes it before requesting the model by name.
Avoid duplicate assistant messages when resuming Cloud threads openai/codex
Resuming a Cloud thread and replaying its history used to re-render completed assistant text when notifications for the same items arrived. Codex now tracks rendered items by turn and item ID and ignores later starts, text deltas, and completions for them, while still accepting unfinished items and new turns.
Include older itemless Cloud failures in full transcripts openai/codex
Failed Cloud turns can carry an error with no items, so paging history through items alone dropped older failures from the transcript. A new FullTranscript scope fetches turn metadata in pages of 100 after item loading and deduplicates overlap while preserving chronological order.
Suppress stale cloud stream disconnects after reopening a thread openai/codex
A queued cloud stream disconnect could outlive a successful reopen and surface as an obsolete reconnect error in the TUI. Losses now arrive as AppServerEvent::ThreadStreamDisconnected and are filtered when a successful attachment to the same thread supersedes them; failed reopens do not.
Improve side-chat discovery and navigation in the TUI openai/codex
The side-chat chooser now shows latest user-prompt previews under saved titles, fetching only for visible entries, with loading and refresh status, an inline retry, and selection preserved across list updates.
