OpenAI Codex 0.157.0 adds GPT-6 Sol and Luna with Amazon Bedrock support, tightens network policy across redirects
OpenAI published Codex 0.157.0 for its Rust CLI on 25 September 2026 at 02:31 UTC. The release adds GPT-6 Sol and Luna with Amazon Bedrock support, enables automatic background-server startup, adds a fork-conversation shortcut, and enforces network restrictions across redirects and live WebSocket traffic.
Why it mattersA team on AWS can now point Codex at GPT-6 Sol and Luna through Bedrock, and network policy changes revoke access to in-flight HTTP and WebSocket traffic instead of only new requests.
A team running Codex on Amazon Bedrock can now use GPT-6 Sol and Luna without switching harnesses. OpenAI published the 0.157.0 release of its Rust CLI on 25 September 2026 at 02:31 UTC, and the changelog names Bedrock support and migration prompts for older models as the first item.
The five features to notice
GPT-6 Sol and Luna are added to the model catalog, and the release notes say Amazon Bedrock is a supported backend for them. A team already on Bedrock for compliance or discount reasons no longer has to run a second Codex install pointed at OpenAI direct.
Fullscreen transcripts are on by default. The previous release put the mode behind /tui; this one turns it on for every launch and adds Shift-click to extend a text selection across the transcript.
Automatic background-server startup runs for eligible interactive sessions. If the local daemon is not already up when Codex starts, the CLI now brings it up on its own. When the server settings do not match the current session, the release notes describe recovery choices rather than a hard failure.
An f shortcut forks a conversation into another app while preserving drafts and queued prompts. Two windows on the same thread now diverge cleanly.
/import works in remote sessions and in local background-server sessions. Before 0.157 it was limited to the plain interactive mode.
The security change to read
The release enforces network restrictions across redirects and ongoing HTTP and WebSocket traffic, and cancels a connection when policy changes revoke its access. On earlier versions a policy update took effect only for new requests, so a long-running agent could keep talking to a host the policy had already denied. Configured proxies now route realtime WebSocket connections and standalone web search, including search redirects. Caller-supplied MITM certificate authorities are supported in the network proxy for teams that inspect their agent's traffic.
What else moved
File upload timeouts run for five minutes instead of one, and transient upload failures are retried. Terminal rendering picks up Unicode bullets, checkboxes, aligned equations and optimization notation. Voice conversations stay open when a user switches threads. Unsent answers move back into the composer when a turn ends. Tmux mouse settings and Terminal.app scrollback over SSH are respected in automatic screen mode. The ultrafast service tier is removed from gpt-5.6-sol.
What it changes for a team on Codex
Three practical shifts. A Bedrock user gets the new models without a second install path. A security lead who ships a network policy update sees the change reach agents that are already running rather than only future ones, so a revoked host is a revoked host. A user who ran into "network policy" errors during search or realtime work will notice those pass on the configured proxy now.
Source
Primary source: openai/codex release rust-v0.157.0 on GitHub, 25 September 2026.
This item was written by an AI system from the linked source. Reveneau is responsible for what it publishes.
Get AI News in your inbox
New developer tools, model and agent releases, and how teams are actually using them to release software. Short, and only when there is something worth reading.
