Codex CLI v0.153.4: GPT-6-Astra Becomes the Bundled Default#

Part of the Codex CLI Version Tracker series. | Codex on GitHub | Official Changelog

Between v0.152.0 and v0.153.4, Codex CLI rolled GPT-6-Astra from an API-only option into the bundled default model, added Vim undo and redo to the composer, gave the plugin CLI remote-marketplace install and removal, and made automatic recaps optional.

GPT-6-Astra, staged across four releases#

The one model that appears new in this range is gpt-6-astra, and it arrived in steps rather than all at once. v0.153.1 let you configure GPT-6-Astra through the API without changing the default model or showing it in the picker.^[3] v0.153.2 corrected its Fast tier description to read "2x speed, increased usage," a display-text change that does not alter how requests run.^[4] v0.153.3 added Astra to the Amazon Bedrock model picker for the Mantle and Runtime global and US routes, and fixed its guidance for asynchronous clarification questions to use the supported tool and accept text only.^[5] v0.153.4 then showed Astra in the bundled model picker and made it the bundled default when no model is explicitly configured, and qualified the async-question guidance by whether the tool is available in the session.^[6]

Features in v0.153.0#

The feature-carrying release in this window is v0.153.0.^[2] The TUI composer gained Vim undo (u) and redo (Ctrl+R), preserving complete drafts including pasted content and attachments. The plugin CLI can now list, install, and remove plugins from remote marketplaces. Setting tui.auto_recap = false disables automatic recaps while keeping manual /recap available. TUI history renders complete patches, the input sent to background terminals, and individual completed commands. Plus and Team users get an earlier warning when less than half of their allowance remains in an approximately five-hour usage window.

Two config surfaces changed alongside those. tui.disable_paste_burst replaces the top-level setting, which stays supported as a fallback. A disabled-by-default features.context_management.experimental_mode was added; when enabled for eligible ChatGPT Plus, Pro, or Pro Lite sessions on the Codex backend, it activates token-budget context, history notes, and a new_context tool, while API-key sessions, custom providers, and temporary structured threads stay excluded.

On the safety layer, Guardian review history now survives compaction, restarts, and user-created forks; Full Access skips Guardian reviews for confirmation-only actions; and User approval mode skips background Guardian scoring and prewarming while retaining sensitive-action checks. v0.152.1 had already made Guardian approval review honor Node REPL policies supplied through model metadata.^[1]

Inside the release#

The span from rust-v0.152.0 to rust-v0.153.4 is 106 commits from 38 distinct authors, with the window running August 31 to September 4, one feature release (0.153.0) followed by a four-step Astra rollout chain. Contributors are almost entirely staff, led by jif-oai (15), bc-openai (14), and etraut-openai (8), with -oai and -openai handles filling most of the list and a short external tail. Work stays concentrated in the Rust workspace, with 284 of 300 file changes under codex-rs and about 10,800 lines added against 1,920 removed.

What this tells us#

The Astra sequence shows how OpenAI ships a new model generation into the terminal: API-only first, then a cloud provider catalog (Bedrock), then the picker, then the default. Each step is a separate point release, so a model reaches "bundled default" only after its routing, tier descriptions, and tool guidance are in place.

The disabled-by-default context_management.experimental_mode flag activates token-budget context, history notes, and a new_context tool when enabled for eligible ChatGPT Plus, Pro, or Pro Lite sessions on the Codex backend. API-key sessions, custom providers, and temporary structured threads stay excluded.

Sources#

This analysis is based on publicly available open-source code and release metadata, conducted for educational and research purposes. All trademarks and software referenced belong to their respective owners.

Let's Connect

Bluesky LinkedIn Threads