$ cat openai/week/2026-09-21.log
the week in review · Sep 21 – Sep 27, 2026
Codex 0.157.0 ships GPT-6 Sol and Luna
By RepoJournal · composed from the cited sources · human-reviewed weekly · methodology
openai-node 7.22.0 and 7.23.0 add the same model identifiers plus GCP external storage, while a required shell call_id change lands and is reverted.
0.157.0 openai/codex
Migration prompts for older models ship in the same release, so a Codex upgrade is not a drop-in for anyone pinning a previous model. The release also turns fullscreen transcripts on by default and enables automatic background-server startup for eligible setups.
v7.23.0 openai/openai-node
The TypeScript SDK's second release of the week adds external storage configuration management, matching the support already present on the Python side. Certificate configuration for external storage also lands, with Azure and AWS client dependencies bumped.
v7.22.0 openai/openai-node
The first of the week's two openai-node releases wires the GPT-6 Sol and Luna identifiers into the generated API surface, roughly a day ahead of the Python SDK. A follow-up fix preserves gpt-5.1-mini and existing audio model choices, which the notes say does not imply model availability.
release: 3.17.0 (#3904) openai/openai-python
The Python SDK's 3.17.0 adds external storage configuration management, plus bindings for conversations, group members, billing, and pin types. pypdf and realtime dependencies move alongside it, and an audio transcription endpoint validator is fixed.
Enforce network policy throughout HTTP and WebSocket requests openai/codex
Destination restrictions and policy revocation now stay in force through redirects, response body reads, and established WebSocket traffic, and denials survive error handling so callers cannot retry or report them as allowed. A companion change adds an optional network.mitm_ca with certificate_file and private_key_file for trusted proxy setups.
Allow subagents to request MCP elicitation input openai/codex
MCP requests needing user input previously failed inside subagents with guidance to ask the parent agent, blocking browser sign-in, form input, and interactive tool approval in the child thread. The root-only restriction is removed, alongside a change that defers subagent activity until the parent answer stream finishes so it no longer flushes the authoritative message.
Move fullscreen transcript control to TUI configuration (#46849) openai/codex
Fullscreen transcript now defaults to false under tui.fullscreen_transcript instead of being on, and --no-alt-screen and tui.alternate_screen = "never" still take precedence. The TUI usage view also gains mouse input: clickable tabs, report controls bound to the same actions as the keyboard shortcuts, and three-line wheel scrolling.
Revert "fix(api)!: declare required local shell output call IDs (#3923)" (#3929) openai/openai-python
The breaking change requiring local shell output call_ids shipped before its source-of-truth changes did, and the revert restores the preceding SDK definitions and generation metadata. Order matters before the tracking ids: 3.17.0 is the branch to stay on this week. Alongside: httpx2's published minimum rises to 2.12.0 for incremental response decompression and stream cleanup on decoding errors...
$ ls openai/week/ # the briefings behind this review
Keep up with OpenAI in about 3 minutes: what actually shipped — the commits, pull requests, releases, and security advisories that matter.
fair warning: these emails are deeply technical. diffs, version numbers, CVEs, benchmark deltas. if that's not your idea of a good read, this isn't your newsletter.
Folds into your digest — weekly by default, monthly if you prefer. Unsubscribe in one click.