Adopting GPT 6.1 Terra in Codex CLI and Security 0.7.0 Updates

Ethan Park

Morning, everyone. Before we get into it, thanks to Jellypod for helping make this daily show a reality. Today's topic is a model, but it's one with real workflow consequences. On October 26, OpenAI announced GPT 6.1 Terra, and it's available right now in Codex for Plus, Pro, Business, Enterprise, and Edu users. It's also on Amazon Bedrock and in the API. The ID is lowercase and hyphenated: gpt, 6.1, terra. It replaces GPT 5.6 Terra as the mid tier, efficiency focused model.

Maya

And this connects to our earlier episode on inline model switching in Codex 0.164.0. If you remember, that release let you pass the model, reasoning effort, and permissions directly to the slash model and slash permissions commands. Well, now there's a model worth switching to. You can test Terra immediately, either with the inline model command or from a session preset, without opening the picker at all.

Ethan Park

Right. So let's start with what changed, because the headline numbers are specific. On Terminal Bench 4.0, Terra scores 50.3 percent at max reasoning effort. GPT 5.6 Terra scored 21.5 percent. So that's more than double, on a benchmark that's close to what Codex does all day: multi step command line work.

Maya

Speed is the other half. OpenAI says Terra finishes multi step computer and terminal tasks over three times faster on average than GPT 6 Astra and GPT 6.1 Sol. And the price drop matters for anyone calling the API directly. Standard pricing is cut by 50 percent compared to both GPT 5.6 Terra and GPT 6.1 Sol. That's one dollar per million input tokens, five dollars per million output tokens, and five cents per million cached input tokens. That cached rate is a 95 percent discount, which is where agent loops tend to live, since they resend a lot of the same context.

Ethan Park

And on a Codex subscription, the claim is that your plan allowance stretches significantly farther on Terra than on Sol or Astra. I'd treat that as directional until you watch your own usage meter, because OpenAI hasn't given a single multiplier here.

Maya

Two more specs. The context window is up to 1,000,000 input tokens, with text and high resolution image inputs. And the reasoning controls include low, medium, and max, plus something I think is underrated: reasoning effort set to none.

Ethan Park

Why does none matter? On GPT 6.1 Sol, the sampling parameters temperature and top p were locked. On Terra, if you set reasoning effort to none, those unlock. Developers in the OpenAI community thread confirmed that behavior. So if you have a pipeline that depends on tuned sampling, say a code formatter or a classifier step, Terra is the model where you can get that back.

Maya

Let's get practical. In Codex CLI 0.164.0 or newer, type slash model, space, gpt 6.1 terra, and you're switched. You can add the effort on the same line: slash model gpt 6.1 terra, then reasoning equals low. Or reasoning equals none if you want the fastest path and don't need it to think. That's the exact pattern the inline arguments in 0.164.0 were built for.

Ethan Park

And pair that with presets. 0.164.0 added ten numbered slash presets slots. A sensible setup is to put Terra in slot one as your daily driver, with your workspace sandbox permissions attached, so quick edits, scripts, test writing, and bug fixes are one keystroke away. Then keep Sol or Astra in another slot for the heavy sessions. Switching becomes a habit instead of a menu trip.

Maya

For AWS shops, GPT 6.1 Terra is generally available on Amazon Bedrock. You'd use the standard Bedrock endpoints, configured through the Bedrock setup wizard in Codex, which became enabled by default in 0.164.0.

Ethan Park

Now the caveats. First, OpenAI itself says Sol and Astra remain better for high difficulty architectural refactoring and subtle algorithmic debugging. Terra can hit limits on multi layer dependency graphs. So don't hand it your multi repo migration just because it's cheaper.

Maya

On reliability, there's good news with a condition. On OpenAI's internal error inducing prompt benchmark, low effort error rates fell from 53.6 percent on GPT 5.6 Terra to 16.7 percent. At max reasoning, it's 4.3 percent, close to Sol and Astra. But that's OpenAI's own benchmark, so it's a reason to test, not a guarantee. And the system card addendum notes slight variances on some edge case moderation prompts, hate and extremism tests in particular, which dropped a little relative to the Astra baseline but stayed within what OpenAI calls acceptable production thresholds.

Ethan Park

One practical gotcha. If you're on a client older than 0.163.0 and the catalog is cached, Terra may not show up. A daemon restart or a manual catalog refresh should populate it. Easiest fix is to just update.

Maya

A few smaller things. Codex Security 0.7.0 shipped on October 26. It adds a validate flag, written with two dashes, that runs a supplemental pass after a scan and saves a separate validation dot md report. It makes extra model calls, leaves your findings and fail on severity decision unchanged, and can't be combined with a max cost cap, dry run, or mock.

Ethan Park

That same release has upgrade notes worth reading. External imports are now opt in. Set CODEX_SECURITY_ENABLE_EXTERNAL_IMPORTS equal to 1 before using publish findings or the GitHub advisories import, and only the literal value 1 works. The inventory helpers moved from Python to Node, so if you called those scripts directly, switch to the bundled launcher with its helper flag. And committed diff scans now need a clean, full checkout.

Maya

Last one: Codex CLI 0.166.0 alpha 1 and alpha 2 landed October 27. Alphas, so no need to chase them. The latest stable is still 0.164.0.

Ethan Park

So the takeaway: update to 0.164.0 or newer, run slash model gpt 6.1 terra on a routine task, and put it in a preset slot if it holds up. That's it for today.