Codex version history

The Codex version history lists the OpenAI Codex CLI releases from 0.95.0 (February 2026) to 0.162.1 (9 October 2026), with each release’s npm publish date, its main features, and the changes that break scripts or configuration, such as removed flags and retired models. Dates come from the npm registry; features come from OpenAI’s GitHub release notes.

This page is for the developer or tech lead who owns Codex in a pipeline, a script, or a shared team config. Codex ships a minor release roughly every week. A nightly codex exec job that installs @openai/codex without a version pin can fail one morning with unexpected argument '--full-auto', and nobody on the team changed anything.

What you get from this Codex changelog

• The breaking changes since August 2026 in one table, each with its version, date, and replacement.
• A seven-step upgrade routine that proves a new version works before your CI depends on it.
• Two copy-paste prompts: one finds removed flags in your repository, one writes an upgrade note for your team.
• The release history by month, one line per release, so you can see what arrived between your version and the latest.

Which Codex version is current?

As of 12 October 2026:

Item	Current value	How to check
Codex CLI (npm latest)	0.162.1, published 2026-10-09	codex --version · npm view @openai/codex version
Pre-release channel (npm alpha)	0.163.0-alpha.5 on 2026-10-12; changes almost daily, so do not pin it in CI	npm view @openai/codex dist-tags
TypeScript and Python SDKs	@openai/codex-sdk 0.162.1 · PyPI openai-codex 0.162.1 (versions match the CLI since 0.155.0)	npm view @openai/codex-sdk version
GitHub Action	openai/codex-action@v1 (newest tag v1.13, 2026-10-05). Since v1.13, Linux drop-sudo requires setfacl (the acl package) and ACL support on /run	the action’s tags on GitHub
Default model	GPT-6.1 Sol since CLI 0.159.1 (2026-09-29), at low effort in Codex; GPT-6 Astra is priority 2 in the picker. Since 0.162.0, new TUI threads respect the model default the server sends, so the effective default for a signed-in account is still set server-side	/model in the TUI · codex debug models --bundled

For model prices, context windows, and effort defaults, see the models hub. This page names models only where a release changed what Codex runs.

Which Codex changes break scripts and config?

Rows up to 0.157.1 were checked against the installed CLI, codex-cli 0.157.1, on 26 September 2026. Where the removal version is known, it comes from OpenAI’s release notes. Rows for 0.158.0 and later, and the dated rows at the end of the table, come from OpenAI’s release notes, OpenAI’s Codex documentation, and the stable codex-cli 0.162.1 binary, all read on 12 October 2026.

Change	Since	What breaks	Use instead
codex exec --full-auto removed	0.147.0 (2026-08-07)	error: unexpected argument '--full-auto' found	A permission profile, -c default_permissions=":workspace" (beta); or --approve-for-me; or --sandbox workspace-write. Pick one: do not combine a profile with --sandbox or --approve-for-me; OpenAI says the profile and legacy sandbox systems do not compose
codex mcp-server removed	0.154.0 (2026-09-09)	Codex can no longer run as an MCP server. In 0.157.1 the CLI reads mcp-server as a prompt and tries to open the TUI, so an MCP client that launches it fails with stdin is not a terminal	codex exec, the Codex SDK, or codex app-server (experimental). codex mcp only manages the servers Codex connects to
-a on-failure and -a untrusted rejected on the command line	checked on 0.157.1; re-run on 0.160.1 (2026-10-10) and on 0.162.1 (2026-10-12)	error: invalid value 'on-failure' for '--ask-for-approval'; 0.160.1 prints error: invalid value 'on-failure' for '--ask-for-approval <APPROVAL_POLICY>' and [possible values: on-request, never]; 0.162.1 rejects both values with the same possible-values list	-a on-request or -a never. In config.toml, approval_policy = "on-failure" still parses as on-request
-a placed after exec	checked on 0.157.1; re-run on 0.162.1 (2026-10-12)	codex exec -a never … fails with unexpected argument '-a'	Drop the flag: codex exec already runs with approval never (on-request under --approve-for-me), and a codex -a … exec prefix does not change that
Untrusted projects no longer supply project AGENTS.md	0.150.0 (2026-08-26)	Codex silently ignores your repository’s instructions in a directory you have not trusted	Trust the project before relying on its AGENTS.md, and check the loaded instructions in a first run
Planning tool off by default	0.152.0 (2026-09-01)	Prompts or harnesses that expect update_plan output get none	Plan mode (/plan), or set tools.update_plan.enabled = true
Default model changed to GPT-6 Astra	0.153.4 (2026-09-04)	Cost, speed, and eval baselines shift for every run that does not set model	Pin model in config.toml or pass -m in CI, and re-run your evals before you move the pin
GPT-5.4 retired; GPT-5.2 and GPT-5.4-mini removed from the bundled catalog	2026-08-31 · 2026-09-09	Configs that name them no longer get those models; GPT-5.4 users see “GPT-5.4 is no longer available” and move to GPT-6 Sol	Choose a current model from the models hub
--profile reads $CODEX_HOME/<name>.config.toml	0.134.0 (2026-05-26) made --profile the primary selector	Team docs that tell people to edit [profiles.<name>] describe the legacy form (still parsed in 0.157.1)	One file per profile, layered on the base config with -p <name>
Feature flags steer, tool_search, remote_control, personality retired	stage removed in 0.157.1 codex features list	--enable steer or --disable tool_search has no effect	Nothing to toggle: the behaviour, where it still exists, is built in. The personality setting is deprecated and selects nothing
GPT-5.4 removed from the bundled catalogs	0.158.0 (2026-09-28)	In 0.157.1 GPT-5.4 was still a hidden, retired entry; it is now gone from the catalog. OpenAI retired it from Codex with ChatGPT sign-in on 2026-08-31, and the API is not affected	Choose a current model from the models hub
Terminal input approval on by default	0.158.0 (2026-09-28)	No error. Commands that run with elevated permissions go through terminal input approval by default (write_stdin_approval is stable and on in 0.162.1; it was under development and off in 0.157.1)	Re-run the smoke test from step 5 of the upgrade routine under the approval policy your CI uses
Bundled plugin-creator skill removed	0.159.0 (2026-09-29)	A prompt, runbook, or plugin guide that calls the plugin-creator skill finds no such bundled skill. The 0.162.1 bundle holds imagegen, openai-docs, review-agent, skill-creator, and skill-installer	The bundled skill-creator skill, or the “Build plugins” flow in OpenAI’s docs
tui.prompt_suggestions setting and automatic follow-up prompt suggestions removed	0.159.0 (2026-09-29)	The setting no longer exists, so a shared config.toml that sets it carries a dead key	Delete the key
Default model changed to GPT-6.1 Sol	0.159.1 (2026-09-29)	Cost, speed, and eval baselines shift again for every run that does not set model; Codex runs GPT-6.1 Sol at low effort by default	Pin model in config.toml or pass -m in CI, and re-run your evals before you move the pin. GPT-6 Astra (priority 2 in the 0.162.1 catalog) stays available for the hardest work
/daybreak unavailable by default	0.161.0 (2026-10-07)	Daybreak is opt-in: daybreak = true alone is not enough, and /daybreak does not appear	--enable cli_daybreak or features.cli_daybreak = true (stage: under development in 0.162.1)
API-key model discovery on by default	0.161.0 (2026-10-07)	No error. api_key_model_discovery was under development and off in 0.157.1 and is stable and on in 0.162.1, so a pipeline that logs in with an API key can behave differently from an older pin	Re-run the smoke test with the same login your CI uses
apply_patch keeps CRLF line endings without an opt-in	0.162.0 (2026-10-08)	The apply_patch_preserve_line_endings flag is at stage removed in codex features list because the behaviour is built in	Delete the flag from any --enable list
New TUI threads follow the server’s model and reasoning-summary defaults; the model picker no longer labels the default	0.162.0 (2026-10-08)	The model a signed-in account starts on is the server’s choice unless you override it at launch, so it can differ from the bundled GPT-6.1 Sol	Pass -m or set model in config.toml: an explicit launch override still wins
GPT-5.5 retires from ChatGPT, ChatGPT Work and Codex	2026-10-14, on all plans (OpenAI models page, checked 2026-10-12)	From that date gpt-5.5 is no longer available in Codex; the OpenAI API is not affected. The 0.162.1 catalog still shows retirement_at: null for it, so the server enforces the date, not the binary	A current model: the 0.162.1 catalog’s upgrade prompt for GPT-5.5 points to GPT-6 Sol. See the models hub
Slack integration is now @ChatGPT	OpenAI docs, checked 2026-10-12	OpenAI’s Slack page is titled “Use ChatGPT in Slack”: mention @ChatGPT, and with Cloud delegation enabled, repository work goes to a Codex Cloud task. Cloud delegation requires a published environment shared with the workspace. Runbooks that tell people to mention @Codex in Slack no longer match that page	Mention @ChatGPT in Slack. Linear still uses @Codex
Chat Completions-only providers not supported	OpenAI docs, checked 2026-10-12 (the release that dropped them is not named)	OpenAI’s models page says “Current Codex releases don’t support wire_api = "chat" or a Chat Completions-only endpoint”	A custom provider that supports the Responses API

Newer models also need newer CLIs: GPT-6 Astra needs Codex CLI 0.153.0 or later, and GPT-6 Sol and GPT-6 Luna need 0.155.0 or later. A pinned old CLI cannot select them, so move the CLI pin before the model pin.

The stable codex-cli 0.162.1 binary, run on 12 October 2026, prints the same top-level codex --help as 0.157.1: no top-level flag or subcommand was added or removed, and mcp-server is still not in the list. The traps in this table still hold on 0.162.1 (checked 2026-10-12): -a accepts only on-request and never, codex exec has no -a, and --full-auto is rejected as an unexpected argument. There is still no /loop in Codex 0.162.1 (not in the slash-command enum at tag rust-v0.162.1, not in the binary, and not in OpenAI’s developer-commands reference). mcp_2026_07_28 is still under development and off in codex features list on 0.162.1.

How do you upgrade Codex without breaking CI?

Treat a Codex upgrade like a dependency bump: one pull request, a pinned version, and a smoke test whose result you can read without reading the agent’s output line by line.

1. Record what you run today and what is available.

   codex --version
   npm view @openai/codex version

2. Read the release notes for every version in between. Each release lives at https://github.com/openai/codex/releases/tag/rust-v<VERSION>, for example rust-v0.157.0. Scan for “removed”, “deprecated”, “default”, and “no longer”. The second prompt below does this for you.

3. Search your repository for the removed flags in the table above. The first prompt below does this and proposes the replacements.

4. Install the new version on a branch, pinned to an exact version:

   npm install -g @openai/codex@0.162.1

   On a workstation, codex update also updates an existing install to the latest version.

5. Run the diagnostics and make config drift fail loudly:

   codex doctor --summary
   codex --strict-config exec --ephemeral --sandbox read-only "Reply with the single word OK."

   codex doctor exits with status 1 when any check fails, so a CI step can gate on it. In a runner without outbound network access, the reachability check fails by design; run it where Codex can reach its provider. --strict-config stops on any config.toml field the new version does not recognise, instead of ignoring it.

6. Run your real automation once with structured output. If your job uses codex exec --output-schema schema.json -o result.json, compare result.json against the last known-good run: same keys, same verdicts on the same fixture. A schema-valid answer on a fixed fixture is your pass signal.

7. Bump the pin in CI in the same pull request. The engineer who owns the CI config approves the merge once steps 5 and 6 pass. If the nightly run fails after merge, revert the pin (npm install -g @openai/codex@<PREVIOUS_VERSION>) before you debug.

Replace <PREVIOUS_VERSION> with the version from step 1.

Copy-paste prompt: find removed Codex flags in this repository

Audit this repository for Codex CLI usage that no longer works, or behaves differently, in codex-cli 0.162.1.
Search shell scripts, Makefiles, package.json scripts, Dockerfiles, .github/workflows,
and docs for:
- `codex exec --full-auto` or `--full-auto` anywhere (removed in 0.147.0)
- `codex mcp-server` or MCP client configs that launch it (removed in 0.154.0)
- `-a on-failure`, `-a untrusted`, `--ask-for-approval on-failure|untrusted`
- `codex exec -a ...` or `codex -a ... exec` (exec has no -a and already runs with approval never; drop the flag)
- `--enable`/`--disable` of steer, tool_search, remote_control or personality
- `[profiles.<name>]` tables in config.toml
- `plugin-creator` skill references (bundled skill removed in 0.159.0) and `tui.prompt_suggestions` in config.toml (removed in 0.159.0)
- `wire_api = "chat"` providers (Chat Completions-only endpoints are not supported)
- `gpt-5.5` or `gpt-5.4` model names in Codex configs, prompts and CI (GPT-5.5 retires from Codex on 2026-10-14)
- `@Codex` Slack mentions in runbooks (Slack uses `@ChatGPT`; Linear still uses `@Codex`)
For every hit, show file:line, the exact line, and a proposed replacement.
Prefer a permission profile (-c default_permissions=":workspace") over --sandbox
for new code, and never propose --dangerously-bypass-approvals-and-sandbox.
Do not edit files. End with a table: file, line, current, replacement, risk.

Start Codex with codex --search so it can read the release pages, then paste:

Copy-paste prompt: write the team's upgrade note

We run Codex CLI 0.150.0 and want 0.162.1.
Read the GitHub release notes for openai/codex tags rust-v0.150.1 through rust-v0.162.1.
Write a one-page upgrade note for our team with three sections:
1. Changes that can break our scripts or CI (removed flags, changed defaults,
   model catalog changes), each with the version that introduced it.
2. New features worth adopting, at most five, each with the command or config key.
3. A verification checklist: the exact commands we run before bumping the pin.
Quote the release-note sentence for every breaking change. If a release has no
published highlights, say so instead of guessing. Do not invent version numbers.

Replace 0.150.0 with your version from step 1. The prompt asks for a quotation behind every breaking change, so the reviewer checks each claim against its source sentence instead of trusting a summary.

What changed in Codex, month by month?

Dates are npm publish dates (UTC). Each row lists the release’s headline changes, not every fix; the full notes are on the GitHub release page for that tag. Model and product events appear in the month they happened.

October 2026

Release	Date	Highlights
0.162.1	2026-10-09	Fixes a TUI crash on multi-line asynchronous questions and startup failures when a running background server’s feature settings differ from the CLI defaults
0.162.0	2026-10-08	Managed Git worktree tools (create and list worktrees from trusted projects); /copy and Ctrl+Insert; tui.mouse_scroll_speed; clickable URLs; live web access and remote compaction for custom Responses-compatible providers; new TUI threads respect the server’s model and reasoning-summary defaults; CRLF preserved in apply_patch without an opt-in; Ultrafast service tiers for Amazon Bedrock Astra models
Model	2026-10-08	Ultrafast speed tier for GPT-6.1 Sol in Codex (GPT-6 Astra has it in Codex and ChatGPT Work from the week of 2026-09-28). In Codex it needs the Pro plan at $500 per month or an eligible Enterprise or Edu plan, and uses 8x included usage or 6x credits (OpenAI speed page, checked 2026-10-12)
0.161.0	2026-10-07	/mcp login <name>; codex exec --cyber-access-program (experimental, per turn: standard, daybreak_blue, or daybreak_red) and the TypeScript SDK option cyberAccessProgram; Daybreak made opt-in (/daybreak needs cli_daybreak); API-key model discovery on by default; GPT-6.1 Sol restated as the default in the bundled and Amazon Bedrock catalogs; Bedrock multi-agent V2, Ultra reasoning, and GovCloud regions; microphone and speaker selection for voice
0.160.1	2026-10-05	Fixes the environment variables (SYSTEMROOT, TEMP, TMP) passed to remote stdio MCP servers
0.160.0	2026-10-01	Projectless sessions with workspace defaults; “Show more” in the agent command center; saved permissions restored on resume; opt-in Guardian (auto-review) context features; explicit provider model catalogs treated as authoritative

September 2026

Release	Date	Highlights
0.159.3	2026-09-30	Optional account-security-setup reminders for ChatGPT sign-in
0.159.2	2026-09-30	Fixes the console-window flash on Windows
0.159.1	2026-09-29	GPT-6.1 Sol becomes the bundled default model, in the bundled and Amazon Bedrock Mantle and Runtime catalogs
0.159.0	2026-09-29	Compact welcome screen; plan-review scrolling; Mermaid rendering; .aws protected under writable roots; bundled plugin-creator skill and tui.prompt_suggestions removed
Model	2026-09-29	GPT-6.1 Sol released at DevDay 2026, with near-Astra performance at a lower cost, according to OpenAI’s Codex changelog. GPT-6 Astra Ultrafast reaches Codex and ChatGPT Work in the same week
Product	2026-09-29	DevDay 2026: Codex Cloud is rebuilt around published, reusable cloud environments, and the earlier setup becomes “Codex Cloud (Legacy)”
0.158.0	2026-09-28	Copy-on-select and right-click paste in the fullscreen TUI; codex mcp add --oauth-client-secret for pre-registered OAuth clients; bearer tokens for exec-server WebSockets; terminal input approval on by default for commands that run with elevated permissions; GPT-5.4 removed from the bundled catalogs; sandbox fixes on Windows, Linux, and macOS
0.157.1	2026-09-26	Patch release; OpenAI published no highlights
0.157.0	2026-09-25	GPT-6 Sol and GPT-6 Luna across providers, including Amazon Bedrock, with migration prompts from older models; fullscreen transcripts by default; the background server starts automatically
0.156.1	2026-09-23	GPT-6 Sol and GPT-6 Luna in the model picker; the rate-limit prompt recommends GPT-6 Luna
0.156.0	2026-09-22	Optional fullscreen /tui; voice on by default; /usage; worktree support on by default (a session uses one only when you pass --worktree or run /worktree); /daemon
0.155.0	2026-09-17	Experimental /voice; live reasoning summaries; archive and delete tasks in the agents overview; Touch ID confirmation for MCP requests on Macs; Python SDK versions now match the CLI
Model	2026-09-14	GPT-5.3-Codex-Spark retired: no longer available in the ChatGPT desktop app, the Codex CLI, or the IDE extension (OpenAI models page, checked 2026-10-12)
0.154.0	2026-09-09	GPT-6 Astra in the picker and Bedrock catalogs; experimental --worktree; inline answers while Codex works; codex mcp-server removed
0.153.3–0.153.4	2026-09-04	GPT-6 Astra in the Bedrock picker; GPT-6 Astra becomes the bundled default model
0.153.0	2026-09-03	codex plugin CLI with remote marketplaces; tui.auto_recap = false turns off automatic recaps; earlier allowance warnings for Plus and Team
0.152.0	2026-09-01	Planning tool off by default; rate-limit banners with credit actions; per-tool MCP output_token_limit

August 2026

Release	Date	Highlights
0.151.0	2026-08-29	Grace period for optional MCP server discovery; extensions can rewrite MCP results; nested subagent tokens count toward goal budgets
0.150.0	2026-08-26	@-mention other tasks; Interrupt hooks; untrusted projects no longer supply project AGENTS.md
0.149.0	2026-08-20	codex agents dashboard to search, start, open, rename, and stop tasks; /cd, /pwd, /cwd; codex queue; broader codex doctor checks; SDK accepts max and ultra effort
0.148.0	2026-08-18	/export to Markdown; codex exec fork; estimated credits or cost in /status for eligible workspaces; built-in Amazon Bedrock provider; async hooks and MCP-tool hooks
0.147.0	2026-08-07	Agent Plugins install; --approve-for-me (automatic approval review); opt-in MCP 2026-07-28 protocol, still under development and off in 0.157.1; codex exec --full-auto removed

July 2026

Release	Date	Highlights
0.146.0	2026-07-29	Named and pinned sessions; side conversations; Agent Plugins manifests and workspace plugin publishing; admin controls for in-app updates
0.145.0	2026-07-21	/import of settings, MCP servers, plugins, and sessions from Cursor and Claude Code; experimental Bedrock login; audio input; GPT-5.4 users migrated to GPT-5.6 Terra or Luna
0.144.6	2026-07-18	Refreshed instructions for GPT-5.6 Sol, Terra, and Luna and corrected their context windows to 272,000 tokens
0.144.0	2026-07-09	writes app-approval mode (read-only actions run, writes ask first); usage-limit reset credits now show their type and expiry and let you choose which one to redeem; a warning when Ultra reasoning with high multi-agent concurrency may use allowance quickly
Model	2026-07-09	GPT-5.6 Sol, Terra, and Luna become generally available; they need Codex CLI 0.144.0 or later. The same day, OpenAI moved the Codex desktop app into the ChatGPT desktop app (secondary: Neowin)
0.143.0	2026-07-08	Remote plugins on by default; system proxy (PAC and WPAD) support; codex remote-control pair; MCP tool search on by default
0.142.5	2026-07-01	Patch in the 0.142 line, which added redeeming usage-limit reset credits in /usage and configurable multi-agent delegation (0.142.0, 2026-06-22)

June 2026

Release	Date	Highlights
0.142.0	2026-06-22	/usage shows and redeems usage-limit reset credits; /plugins groups remote plugins into OpenAI Curated, Workspace, and Shared; rollout token budgets across agent threads; multi-agent delegation set to disabled, explicit-only, or proactive; indexed web-search mode
0.141.0	2026-06-18	End-to-end encrypted relay channels for remote executors; executor plugins can start their stdio MCP servers per thread; TUI prompts can auto-resolve after inactivity
0.140.0	2026-06-15	/usage token views; codex delete and /delete; /import from Claude Code; @ opens the unified mentions menu by default; encrypted local storage for CLI and MCP OAuth credentials
0.138.0	2026-06-08	/app hands the current CLI thread to the desktop app on macOS and Windows; saved image paths exposed to the model; plugin commands gain JSON output
0.137.0	2026-06-04	F13–F24 keybindings; codex remote-control controller pairing; parallel standalone web search
0.136.0	2026-06-01	/archive, codex archive, and codex unarchive; clickable links in TUI Markdown; alpha codex sandbox setup --elevated for Windows admins

May 2026

Release	Date	Highlights
0.135.0	2026-05-28	Richer codex doctor diagnostics; /status shows remote connection details; Vim text objects; /permissions lists named permission profiles
0.134.0	2026-05-26	Search across local conversation history; --profile becomes the primary profile selector; per-server environment targeting and OAuth options for MCP; read-only MCP tools run concurrently when they set readOnlyHint
0.133.0	2026-05-21	Goals on by default; codex remote-control runs in the foreground; permission profiles gain inheritance and managed requirements.toml support
0.132.0	2026-05-20	Python SDK authentication; codex exec resume accepts --output-schema
0.131.0	2026-05-18	codex doctor; one @ picker for files, directories, plugins, and skills; Python SDK renamed to openai-codex
0.130.0	2026-05-08	codex remote-control for a headless app server; plugin sharing controls; thread pagination for app-server clients; Bedrock auth from aws login profiles
0.129.0	2026-05-07	Modal Vim editing (/vim); redesigned resume and fork picker; browse and toggle hooks in /hooks; theme-aware status line

April 2026

Release	Date	Highlights
0.128.0	2026-04-30	Persistent /goal workflows; codex update; permission profiles with built-in defaults; plugin marketplace install and plugin-bundled hooks; import of sessions from external agents; explicit multi-agent v2 limits
0.124.0	2026-04-23	Alt+, and Alt+. lower and raise reasoning effort; Amazon Bedrock support with AWS SigV4 signing; hooks stable and configurable in config.toml and requirements.toml; Fast service tier by default on eligible ChatGPT plans
Model	2026-04-23	GPT-5.5 (gpt-5.5-2026-04-23) released; Codex now labels it a legacy model
0.122.0	2026-04-20	/side conversations; Plan mode shows context usage before implementation; deny-read filesystem globs; codex exec --ignore-user-config and --ignore-rules for isolated runs
0.121.0	2026-04-15	codex marketplace add (now codex plugin marketplace add); Ctrl+R prompt history search; memory reset and deletion controls; secure devcontainer profile

February and March 2026

Release	Date	Highlights
0.117.0	2026-03-26	Plugins become a first-class workflow (/plugins, startup sync, install and remove); subagents get path-based addresses such as /root/agent_a
0.98.0	2026-02-05	GPT-5.3-Codex support; steer mode stable and on by default (Enter sends during a running task, Tab queues)
0.95.0	2026-02-04	codex app <path> launches the desktop app on macOS; personal skills from ~/.agents/skills; shell tools run in parallel

What breaks after a Codex upgrade, and how do you recover?

• A CI job fails with unexpected argument. A flag was removed or moved. Match the message against the breaking-changes table, apply the replacement, and re-run the smoke test from step 5. If the fix is not obvious within the hour, revert the pin and fix it in a branch.
• An MCP client hangs or reports stdin is not a terminal when it starts Codex. The client still launches codex mcp-server, which 0.154.0 removed. Drive Codex through codex exec or the SDK instead, and delete the server entry from the client’s config.
• The agent ignores AGENTS.md after the upgrade. Since 0.150.0, an untrusted project supplies no project instructions. Trust the directory, then ask Codex in a fresh session to restate the repository rules it loaded; a missing rule shows up in the first answer.
• Costs or run times jump with no config change. The default model moved (GPT-6 Astra from 0.153.4, GPT-6.1 Sol from 0.159.1). Pin model explicitly, compare a week of /status or /usage numbers, and change the pin only after your evals pass.
• codex --strict-config fails on a key you have used for months. The new version no longer recognises it. Look the key up in the release notes, move it to its replacement, or delete it if it was a retired feature flag.
• A model name fails on an older CLI. GPT-6 models need 0.153.0 (Astra) or 0.155.0 (Sol and Luna). Upgrade the CLI first; do not work around it by running with a model the team has not evaluated.
• A task that names gpt-5.5 fails in Codex. GPT-5.5 retires from Codex on 2026-10-14 (OpenAI models page, checked 2026-10-12). Move the pin to a current model from the models hub and re-run your evals first.
• A custom provider stops working. Current Codex releases do not support wire_api = "chat" (OpenAI models page, checked 2026-10-12). Point Codex at a provider that supports the Responses API.

Where to go next after a Codex upgrade

Codex reference hub The other Codex reference pages

Install and update Codex Install methods and first-run checks

Codex configuration config.toml, profiles, and permission profiles

Run Codex non-interactively codex exec, --output-schema, and CI gates

Codex in CI/CD The GitHub Action and pinned pipelines

Models hub Current models, prices, and defaults across tools

The sibling changelogs track the other two tools the same way: Claude Code version history and Cursor version history. For the full, unabridged notes, read the Codex releases on GitHub.