This feeling is addictive
Categories:
Half a year after the release of the GPT5 series, its logical rigor leads the field, but the most criticized aspect is its slow speed, especially GPT5.2, which is absurdly slow. For many ordinary requests, opus may start making changes after 20 seconds of context gathering, while GPT5.2 might analyze for 20 minutes before starting to make changes. The concurrent GPT5.2 codex model is slightly faster, but its logic is noticeably weaker. Later, the GPT5.3 codex model balanced logic and speed, and the codex model became usable. GPT5.4 integrated coding capabilities into the main model, no longer releasing a separate codex model, and its speed was significantly improved. It is said that the effective rate per token increased, so it can think less while achieving a similar accuracy.
Those who have seen the LLM thinking chain will notice that, even when the conclusion is correct, not all thoughts in the chain are correct—quite a few are erroneous. GPT5.4 reduces ineffective thinking to speed up, which I personally interpret as the model becoming smarter.
At the same time, the codex cli also opened a fast mode, which is estimated by the community to consume double the quota at 1.5x the original speed. I recently indulged and turned on fast, and I think the perceived speed is already very close to opus, while possessing stronger logical analysis capabilities.
I also noticed that after enabling fast, the two five-hour windows may be used up along with the weekly quota, meaning under high load (GPT5.4 fast), the plus/team weekly quota can be exhausted in a single day.
Previously I recommended cliproxyapi, specifically for switching codex accounts. Recently I found its API implementation incomplete, so I developed my own account switcher to exploit the team loophole and enjoy it for a while.
Open source at: github.com/jqknono/codex-account-switch Quick usage: npx codex-account-switch
