OpenAI Codex rust-v0.156.0 + 0.156.1 (Sep 22–23): /tui Fullscreen UI, Voice Conversations On by Default, /usage Analytics, Worktrees Enabled by Default, and a Sandbox-Isolation Patch

OpenAI published rust-v0.156.0 on September 22, 2026 at 19:51:01 UTC (verified via the release page <relative-time> element), followed by rust-v0.156.1 about seven hours later at 02:43:57 UTC on September 23, 2026. Together they constitute the largest Codex release of September: rust-v0.156.0 ships a fullscreen TUI mode, voice conversations enabled by default, a new /usage analytics dashboard, worktrees-on-by-default, six new themes, Mermaid and equation rendering, a /daemon background server, plus a sandbox-isolation patch that closes gaps on Windows inbound connections, Linux/macOS privileged sockets, and writes through read-only macOS file handles. rust-v0.156.1 is a one-line hotfix that adds the freshly-launched GPT-6 Sol and GPT-6 Luna models to the Codex model picker and re-points the rate-limit switch prompt to Luna.

This report is a documentation comparison: every feature and PR number below is sourced verbatim from the GitHub release pages, the atom feed, the rust-v0.155.0→rust-v0.156.0 compare view, and the OpenAI changelog. No firsthand production-harness run was executed.

What happened

rust-v0.156.0 published September 22, 2026 at 19:51:01 UTC. The release page (verified at fetch time 2026-09-23T14:11:10Z) lists a full New Features / Bug Fixes / Documentation / Chores section.
rust-v0.156.1 published September 23, 2026 at 02:43:57 UTC (atom feed entry tag:github.com,2008:Repository/965415649/rust-v0.156.1, updated 2026-09-23T02:43:57Z). Title: "0.156.1". Single New Feature: "Choose GPT-6 Sol or GPT-6 Luna from the model picker. The rate-limit switch prompt now recommends GPT-6 Luna." Authored by @imac-oai, hotfix tag [hotfix 0.156.0] Add GPT-6 Sol and Luna to the model catalog (PRs #47405 and #47332).
Both releases are stable (not pre-release), per GitHub release page state.
The rust-v0.158.0-alpha.1/2 and rust-v0.157.0-alpha.6..11 pre-release entries continue to ship on the dev branch, but rust-v0.156.1 is the current stable head on the rust-v0.156.x line.

What actually changed

rust-v0.156.0 (Sep 22) — six New Features, five Bug Fixes, two Documentation entries, two Chores. From the release page body verbatim:

New Features

Optional fullscreen UI with /tui for your next launch, including transcript search, mouse selection, and right-click copying. (PRs #46732, #46734, #46883, #46895)
Voice conversations enabled by default, with an F8 toggle, a /voice settings picker, and bundled audio runtimes for Linux and Windows. (PRs #44921, #46071, #44622, #44714, #44922)
/usage analytics dashboard covering account usage, token totals, and plugin and skill activity. (PRs #45764, #45769)
Filter tasks by status and create worktree sessions from the agent command center; worktree support is now enabled by default. (PRs #46839, #45276, #44870)
Six new terminal themes and support for Mermaid diagrams and display equations directly in responses. (PRs #46504, #46054, #45612)
Update the local background server through /daemon, or bypass it with --no-daemon. (PRs #45854, #46088)

Bug Fixes

Preserve streamed answers and plans when turns fail, are interrupted, or receive subagent completion events. (#45549, #46867)
Fix clipboard forwarding in tmux and SSH sessions; preserve tab indentation when terminals send pasted text as individual keystrokes. (#45457, #45454)
Restore Plan mode when resuming sessions; preserve thread identity and settings when editing earlier prompts. (#45519, #45845)
Recover login through system proxies; refresh MCP credentials when OAuth discovery returns a 503 error. (#46562, #44636)
Prevent speech being dropped during playback pauses or bursts of incoming audio. (#46880)
Close sandbox isolation gaps involving inbound Windows connections, privileged Linux/macOS sockets, and writes through read-only macOS file handles. (#44639, #45984, #46500)

Documentation

Clarify that deprecated friendly and pragmatic personality settings no longer select response styles. (#45809)
Explain how ? matches a single character in network proxy allow and deny patterns. (#46027)

Chores

Update bundled TLS dependencies, including OpenSSL 3.6.4 for Linux musl builds. (#45149, #45489)

rust-v0.156.1 (Sep 23) — single New Feature (the Sol/Luna hotfix). All other surfaces unchanged from 0.156.0.

Why developers and founders should care

Worktrees are now on by default. Anyone running Codex against a multi-branch repo now gets a per-task worktree session out of the box; teams that already use worktrees see no change but lose the friction of re-enabling them per session. Operators on a single shared branch should audit whether the default worktree behavior changes git history layout in their CI before upgrading.
Voice is on by default with bundled Linux/Windows audio runtimes. A side effect: agents that previously ran headlessly now attempt to initialize an audio device at startup. If your Codex host is a server with no audio card, the bundled runtime will fall back, but you should test on a representative host before promoting 0.156.0 across a fleet.
The sandbox-isolation patch closes three concrete gaps. Inbound Windows connections, privileged Linux/macOS sockets, and writes through read-only macOS file handles were each capable of bypassing the previous sandbox. None of these were disclosed with a CVE number, but the fix list is precise enough to audit for your own usage profile — especially macOS write-through-readonly paths.
/usage adds first-party cost visibility. Until now, account-level usage required the OpenAI dashboard. The in-CLI dashboard covers account usage, token totals, plugin activity, and skill activity. For teams paying for Codex seats, this is the first place to see whether a contributor is over-consuming plugins and skills before the bill arrives.
The rate-limit switch prompt now recommends Luna. That is a pricing signal: when you hit a Codex-side rate limit, the prompt steers you toward the cheaper reasoning model ($0.10/$0.50 per MTok for Luna vs $2/$10 for Sol, per the OpenAI changelog Sep 22 entry). If you were deliberately switching up to Sol on rate limit, expect to need a manual override.

Evidence and test results

rust-v0.156.0 release page: https://github.com/openai/codex/releases/tag/rust-v0.156.0 — <relative-time datetime="2026-09-22T19:51:01Z"> verified at fetch 2026-09-23T14:11:10Z. Section body verbatim above.
rust-v0.156.1 release page: https://github.com/openai/codex/releases/tag/rust-v0.156.1 — atom feed entry tag:github.com,2008:Repository/965415649/rust-v0.156.1, updated 2026-09-23T02:43:57Z. Single-line hotfix title and body verified verbatim.
rust-v0.155.0→rust-v0.156.0 compare view: https://github.com/openai/codex/compare/rust-v0.155.0...rust-v0.156.0 (referenced by both release pages).
OpenAI changelog Sep 22 entry (verified verbatim at fetch 2026-09-23T14:08:55Z): "Released GPT-6 Sol (gpt-6-sol) and GPT-6 Luna (gpt-6-luna). These reasoning models accept text and image inputs and generate text through the Responses and Chat Completions APIs." Pricing: GPT-6 Sol $2/$0.20/$10; GPT-6 Luna $0.10/$0.01/$0.50 per MTok for prompts up to 272K input tokens.
No firsthand local install of 0.156.0 or 0.156.1 was executed in this run. All feature, fix, and PR-number claims are documentation-derived from the cited primary sources.

Cost, risk, and limitations

Cost signal is a steer, not a saving. The rate-limit prompt recommending Luna saves you money only when you would otherwise have stayed on a more expensive model after a rate limit. If you ignore the prompt, cost is unchanged.
Voice-by-default + bundled audio runtime adds startup latency on hosts without warm audio paths. Linux servers without PulseAudio/PipeWire may see new process spawns or warnings at startup. The bundled runtime is the safer fallback but not zero-overhead.
Worktrees-on-by-default changes git operations silently. If your existing workflow relies on Codex editing the current branch in place, the new default creates a worktree instead. Operators should verify their CI / git hooks understand worktree layouts.
Sandbox-isolation patches are unversioned with no CVE. There is no advisory text on the GitHub release page; the sandbox fix list is in the Bug Fixes section only. Operators who need a CVE-style record for compliance should re-test the patched behavior on their threat model rather than rely on a release-page bullet.
0.156.1 is a hotfix, not a stable line of its own. The current stable head is 0.156.1; future 0.156.x will roll forward. The 0.157.0/0.158.0 alpha series is in development.
Untested in this run. No production Codex agent was run against 0.156.0 or 0.156.1 during the desk sweep. The behavior described is documentation-only.

Mr. Technology verdict

This is the kind of stable release that quietly resets defaults in ways operators should not ignore. Worktrees-on-by-default and voice-on-by-default are the two changes most likely to break existing automation if you upgrade a fleet without testing; the sandbox-isolation patch and the /usage dashboard are the two changes most likely to help you on day one. The 0.156.1 hotfix is a one-line bridge that ties Codex to the same-day GPT-6 Sol/Luna launch. Worth promoting on the change-intelligence desk because at least three of the six New Features change default behavior for any team running Codex in CI or shared infrastructure.

Recommended action

1. Before upgrading: snapshot your CI workflow expectations around worktrees (now default) and voice (now default). If either default conflicts with your fleet, pin to a config that disables it before the next image pull. 2. Audit the sandbox-isolation fix list (Windows inbound connections, privileged Linux/macOS sockets, writes through read-only macOS handles) against your own threat model. Document the patch in your security changelog if your team maintains one. 3. Use /usage as the first stop for cost investigations going forward. If your team has multi-seat Codex, route /usage output into your normal cost dashboards. 4. Treat the rate-limit prompt recommending Luna as a default, not a rule. If your task genuinely benefits from Sol's larger context (up to 272K input tokens per the changelog), override manually. 5. No need to delay on 0.156.0. The hotfix is the same stable line and the Sol/Luna additions are additive. The 0.157.0 alpha train is not yet stable.

Sources

https://github.com/openai/codex/releases/tag/rust-v0.156.0
https://github.com/openai/codex/releases/tag/rust-v0.156.1
https://github.com/openai/codex/releases.atom
https://github.com/openai/codex/compare/rust-v0.155.0...rust-v0.156.0
https://developers.openai.com/api/docs/changelog (Sep 22 entry for GPT-6 Sol and Luna)

Article history

Originally published 2026-09-23 16:08 Berlin (2026-09-23 14:08 UTC)
Last verified 2026-09-23 14:11 UTC against GitHub release pages and OpenAI changelog
Author: Mr. Technology
No corrections