Codex CLI 0.157.0 Adds GPT-6 Sol and Luna

Codex CLI 0.157.0 adds GPT-6 Sol and GPT-6 Luna to the bundled model support, including Amazon Bedrock routes and migration prompts for older models. Released September 25, 2026, it also changes interactive-session startup, transcripts, imports, thread forking and network enforcement. Install the exact release with npm install -g @openai/codex@0.157.0. Read the official Codex changelog.

Codex CLI 0.157.0 model support

The model update turns the broader September 22 rollout into a concrete client release. Codex CLI 0.156.1 had already exposed Sol and Luna in the model picker and made Luna the recommended rate-limit switch. Version 0.157.0 incorporates GPT-6 Sol and Luna support into the larger release, adds Amazon Bedrock support and includes migration prompts for older models.

Use /model inside the CLI or launch a task with codex --model gpt-6-sol or codex --model gpt-6-luna. Sol is positioned for complex coding and agent work, while Luna is the lower-cost choice for focused, high-volume tasks. Our GPT-6 Sol and Luna launch report covers the model roles, published API prices and plan-level rollout. This release is the dated client implementation update.

Installing 0.157.0 does not by itself grant model access. OpenAI says availability can depend on the account rollout, plan, workspace settings and Enterprise administrator controls. If the version is current but a model is absent or rejected, verify the signed-in account and workspace policy before treating the installation as failed. The site’s GPT-6 Sol model page and GPT-6 Luna model page keep the model facts separate from this CLI release.

Migration prompts are also different from an automatic model replacement. The changelog says the release adds prompts for older models; it does not announce that every existing configuration is rewritten. Preserve the configured model, provider and reasoning effort when upgrading, then inspect the model picker and run one representative task before changing a shared default.

GPT-6 Sol and Luna on Amazon Bedrock

The Bedrock addition matters because OpenAI’s first-party API IDs and Bedrock identifiers are not interchangeable. OpenAI’s Bedrock guide says GPT-6 Sol and Luna are available through both Bedrock Runtime and Mantle. For Mantle in us-east-1, the model IDs are openai.gpt-6-sol and openai.gpt-6-luna.

Bedrock Runtime uses inference profile IDs. The United States profiles are us.openai.gpt-6-sol and us.openai.gpt-6-luna; the global profiles are global.openai.gpt-6-sol and global.openai.gpt-6-luna. The guide lists Sol and Luna for Mantle in us-east-1 in North Virginia, while GPT-6 Astra uses us-west-2 in Oregon. Check OpenAI’s Amazon Bedrock guide.

That region difference is easy to miss because examples on the documentation page still use GPT-5.6 Sol in us-east-2. OpenAI explicitly instructs users trying GPT-6 Sol or Luna to change both the model ID and the Region. Copying the example and changing only the model string can therefore leave the request pointed at an unsupported route. Ruby users also need to update the Region embedded in an explicitly configured base URL.

The Bedrock guide distinguishes Runtime from Mantle in another practical way: background mode is available through Mantle subject to data-retention settings, but not through Bedrock Runtime. Both routes support streaming responses, and computer use is available on supported models. Choose the route against the workflow requirements rather than assuming the provider label alone determines equivalent behavior.

Background sessions, imports and thread forks

Codex CLI 0.157.0 enables automatic background-server startup for eligible interactive sessions. When existing server settings are incompatible, the client presents recovery choices instead of silently assuming the background path is usable. This builds on the previous release’s /daemon control and --no-daemon bypass. Teams with pinned launch scripts should verify whether their session now starts the server automatically and whether their explicit opt-out still reflects the intended workflow.

The release also makes /import available in remote sessions and local background-server sessions. An imported task or context can now enter workflows that are not confined to a conventional foreground local session. The useful acceptance check is to confirm what was imported, where the resulting task is running and which repository or environment it can access.

A new f shortcut forks a conversation that is open in another app while preserving drafts and queued prompts. This is a continuity feature: it lets a user branch work without discarding text already prepared for the original conversation. After forking, verify which thread receives the next prompt so a queued instruction is not mistaken for work applied to both branches.

Fullscreen transcripts are now enabled by default. Shift-click extends text selections, and terminal rendering gains Unicode bullets, checkboxes, aligned equations and optimization notation. These changes affect how evidence and long outputs can be reviewed in the terminal; they do not alter model reasoning or API pricing.

Network and recovery fixes to verify

The release fixes configured proxy routing for realtime connections and standalone web search, including search redirects. It also enforces network restrictions across redirects and ongoing HTTP and WebSocket traffic, with cancellation when a policy change revokes access. This is a meaningful boundary change for managed environments: a redirect should not escape the active policy, and a connection should not remain authorized after the policy is tightened.

File uploads now retry transient failures and allow up to five minutes before timing out. The client recovers unsent answers to interactive questions into the composer when a turn ends, while preserving active history searches. Voice calls remain active when switching threads, and Terminal.app over SSH regains native scrollback in automatic screen mode while respecting tmux mouse settings.

A focused post-upgrade check should cover the path a team actually uses. Confirm codex --version, open /model, start one background-eligible session and test the relevant proxy or upload route. For code work, the GPT-6 Sol coding guide supplies a bounded review-and-test workflow; it does not replace the client and provider checks.

The material update is not another model launch or a pricing change. Codex CLI 0.157.0 is the client release that bundles Sol and Luna support across first-party and Bedrock configurations while changing several terminal and background-session behaviors. Upgrade when those capabilities matter, then validate the exact account, provider, Region and session mode that your work depends on.