Will we be able to use GPT-5 Pro in codex?

this is what I use to get reasoning effort from medium to high ,
codex --ask-for-approval never --sandbox danger-full-access -c model_reasoning_effort="high"

this is what I use to get reasoning effort from medium to high , codex --ask-for-approval never --sandbox danger-full-access -c model_reasoning_effort="high"

Thanks for the command, what are the limits using "high" reasoning, I hit a limit after 1-2 hours of using it on Plus plan. Do you have an idea? @Aze1338 ty

> this is what I use to get reasoning effort from medium to high , codex --ask-for-approval never --sandbox danger-full-access -c model_reasoning_effort="high" Thanks for the command, what are the limits using "high" reasoning, I hit a limit after 1-2 hours of using it on Plus plan. Do you have an idea? @Aze1338 ty

That's exactly hte limit. On plus plus plan i heard from the official doc that is around 1-2 hours per week.

What about Pro plan?

Pro has a higher limit that was also bumped up this afternoon according to this tweet: https://x.com/embirico/status/1953981033122345342

Sounds like we might get in-app info next week.

If you were already using the CLI my understanding is you might need to logout and then re-login and select sign in instead of API

Pro actually has near unlimited usage with Codex CLI - as long as you don't script it/multi account it.

Pro actually has near unlimited usage with Codex CLI - as long as you don't script it/multi account it.

That is not true. Codex dies in an hour or two of hard work.

> Pro actually has near unlimited usage with Codex CLI - as long as you don't script it/multi account it. That is not true. Codex dies in an hour or two of hard work.

Absolutely wrong, I've been intensively using it with high reasoning effort for the entire day without issues.

Absolutely wrong, I've been intensively using it with high reasoning effort for the entire day without issues.

Absolutely true as of Aug 9. They bumped limits soon after.

hey folks, at the moment we don't have plans to enable using GPT-5 Pro in Codex

hey folks, at the moment we don't have plans to enable using GPT-5 Pro in Codex

What's the reason behind this?

I'd like to petition for making GPT-5 Pro access available in Codex CLI. there could be a plan mode where GPT-5 Pro is available for. Or have a % split of GPT-5 Pro available as how Claude Code does it. If Anthropic can make the expensive Opus 4.1 model available in Claude Code Max tiers, I don't see how OpenAI can't afford it with % access to GPT-5 Pro? Below are some more details on the why:

Why: Pro-tier planning would reduce hallucinations and improve spec/plan quality for complex R&D work (climate, finance, large simulations).

Priors: Aider’s architect+editor results and OpenHands’ planner/coder requests show the benefit of a stronger planner model.
Aider
GitHub

Business case: Unlocking GPT-5 Pro in Codex would deepen loyalty among Pro subscribers who value raw model capability over harness alone. It'd make Codex CLI another round of YoutTuber attention, which will help broader audience realize the new Codex CLI is actually better than Claude Code or Gemini CLI when it comes research-grade agentic dev.

Please! Pro is way better than Claude code opus.

I purchased the PRO version, but the Codex "/model" parameter does not have a GPT-5 PRO option. This is a huge disappointment.

I purchased the PRO version, but the Codex "/model" parameter does not have a GPT-5 PRO option. This is a huge disappointment.

For real

hey folks, at the moment we don't have plans to enable using GPT-5 Pro in Codex

Claude has noticeably declined over the past three weeks - not broken, but the quality has dropped. MCP, CLI have become less intuitive, the model thinks in templates, lost its individuality. It solves tasks worse, and sessions hit limits faster. People aren't panicking, but many are starting to look at alternatives. What's the point of a $200 Pro subscription if GPT-5 Pro isn't available in Codex CLI? We work in the terminal, not sitting in a browser. Paying 10x more and not getting the best model where it's actually needed - strange logic. Enable GPT-5 Pro in Codex CLI for Pro subscribers. This isn't just a feature - it will show that the Pro tier is actually worth the money. The timing is perfect now, while Claude isn't fixed and while developers are open to switching.
The window of opportunity is open, don't miss it.

> I purchased the PRO version, but the Codex "/model" parameter does not have a GPT-5 PRO option. This is a huge disappointment. For real

Interestingly, when you launch it with --model gpt-5-pro, it reports to use this model:

❯ codex --model gpt-5-pro

╭────────────────────────────────────────────────────────╮
│ >_ OpenAI Codex (v0.39.0)                              │
│                                                        │
│ model:     gpt-5-pro high   /model to change           │
│ directory: ~                                           │
╰────────────────────────────────────────────────────────╯

  To get started, describe a task or try one of these commands:

  /init - create an AGENTS.md file with instructions for Codex
  /status - show current session configuration
  /approvals - choose what Codex can do without approval
  /model - choose what model and reasoning effort to use

> > I purchased the PRO version, but the Codex "/model" parameter does not have a GPT-5 PRO option. This is a huge disappointment. > > For real Interestingly, when you launch it with --model gpt-5-pro, it reports to use this model: `` ❯ codex --model gpt-5-pro ╭────────────────────────────────────────────────────────╮ │ >_ OpenAI Codex (v0.39.0) │ │ │ │ model: gpt-5-pro high /model to change │ │ directory: ~ │ ╰────────────────────────────────────────────────────────╯ To get started, describe a task or try one of these commands: /init - create an AGENTS.md file with instructions for Codex /status - show current session configuration /approvals - choose what Codex can do without approval /model - choose what model and reasoning effort to use ``

I think that's a bug. I just tried it with gpt-5-ultra, and that worked, too 😆