Please restore 5.3-codex as a selectable model

Summary

Please bring back 5.3-codex as a selectable model.

From real day-to-day coding use, 5.4 feels significantly less balanced than 5.3-codex for practical software work.

Why this matters

For my workload, 5.4 appears much more expensive in terms of usage budget, but the quality does not justify that tradeoff.

Observed problems with 5.4 compared to 5.3-codex:

• average execution time is more than 2x higher on similar coding tasks
• it burns through the Pro allowance much faster
• output quality is not consistently better
• it often feels unstable during longer sessions
• it stops mid-task more often
• it loses context more often
• it forgets the current objective
• it drifts away from the original instruction set
• it sometimes feels like it is reasoning more while delivering less

Concrete impact

I am on a Pro plan. Before this change, I could code productively throughout the week.

With 5.4, I burned roughly 30% of my weekly allowance in a single day.

This would be easier to accept if the results were clearly better, but in my experience they are not. In practice, 5.4 feels slower, less predictable, and less efficient than 5.3-codex.

So the current situation is:

• higher usage cost
• worse budget efficiency
• no reliable quality gain
• weaker session stability

Why 5.3-codex should return

5.3-codex felt much better balanced for real coding work:

• faster response and completion times
• better budget efficiency
• more stable multi-step execution
• better context retention
• better overall value for Pro users

Even if 5.4 remains the default or recommended model, users should still have the option to choose 5.3-codex when they need:

• lower latency
• more predictable behavior
• better cost/usage efficiency
• a model that is more reliable for long coding sessions

Request

Please restore 5.3-codex as an available selectable model, or provide a comparable lower-latency, better-balanced alternative.

At minimum, it would help to have:

• a user-selectable fallback to 5.3-codex
• clearer guidance on which model is optimized for long coding sessions
• transparency around why 5.4 is considered better despite these real-world regressions

Right now the value proposition is hard to understand: paying for Pro makes much less sense if the newer model consumes the quota much faster while also feeling less reliable in actual coding workflows.