Suspected silent model routing: selected 5.5 High now behaves like 5.6 Sol Max

What issue are you seeing?

I want to report a serious regression in Codex/ChatGPT behavior that appeared within the last 24 hours, roughly between August 5 and August 6, 2026.

The UI shows that I have selected 5.5 High. I intentionally use 5.5 High because it used to handle difficult Codex tasks reliably and complete them in a reasonable amount of time, usually around 30-60 minutes.

Over the last day, the behavior of the selected 5.5 High model changed sharply. It no longer feels like the previous 5.5 High behavior. Instead, it behaves much more like 5.6 Sol Max: a newer/higher/heavier model variant that, in my real Codex workflow, performs worse for these tasks.

This is not a complaint that I was moved to a lower model. The concern is the opposite: it feels like I may have been silently routed to a higher or different model/variant, but that model is substantially worse for my use case.

Observed behavior:

• Tasks take much longer than before.
• The model starts going in circles.
• Progress becomes unstable.
• Long sessions can run for 5-6-10 hours without a clean solution.
• The model does a lot of work but fails to actually finish the task.
• Planning and task completion quality are noticeably worse than the previous 5.5 High behavior.
• The UI still says 5.5 High is selected.

I cannot prove that a model substitution or routing change happened, but the observed behavior strongly feels like the selected 5.5 High is no longer the same effective behavior I was choosing before.

This is critical for trust in model selection. If a user explicitly selects 5.5 High because it works better for their tasks, the system should not silently move them to another model, routing path, fallback, capacity mode, or internal variant, even if that other model is considered newer, higher, or more capable in general.

What steps can reproduce the bug?

1. Use Codex/ChatGPT with 5.5 High selected in the UI.
2. Run substantial multi-step coding or maintenance tasks that previously completed reliably on 5.5 High.
3. Compare behavior from before August 5, 2026 with behavior observed on August 5-6, 2026.
4. Observe that the selected model now behaves more like 5.6 Sol Max: slower, more circular, and less able to finish tasks.

Actual behavior:

• The model displayed as 5.5 High behaves like a different, heavier, slower, less effective variant.
• It can spend many hours working without reaching a correct finished state.
• The practical result is a major regression compared with the previous 5.5 High behavior.

What is the expected behavior?

If the UI says 5.5 High is selected, I expect the effective behavior of 5.5 High: stable task execution, reasonable completion time, and the ability to finish real Codex tasks.

If OpenAI changes routing, fallback behavior, capacity handling, degraded mode, or an internal model version behind the same visible model name, that should either be visible to the user or at least not cause a major regression in quality and task completion.

Additional information

Environment/context:

• Surface: Codex / ChatGPT model selection
• Selected model shown in UI: 5.5 High
• Suspected effective behavior: similar to 5.6 Sol Max
• Date observed: August 5-6, 2026

This is not just a subjective complaint about latency. The difference is visible in real Codex work: tasks that previously completed in 30-60 minutes on 5.5 High can now run for 5-10 hours and still fail to complete properly.

Please check whether routing, fallback, capacity handling, or internal model versioning changed in the last 24 hours for users who have 5.5 High selected.