1
Curious builder0 XP earned · 300 to level 2
0 daysFinish a lesson to begin
Badge collection0 of 6 unlocked
26 small wins to finish your pathNext lesson →
Models and fast mode: speed and cost
Three things change what a session costs: which model, how much it reads, and whether you asked for speed.
ChatGPT Work and Codex share pricing, credits and usage limits, so a heavy Codex day shows up in the same place as everything else on that account.
Fast mode
Codex can run the model faster for more credits, and the documentation gives the numbers. For GPT-5.6, GPT-5.5 and GPT-5.4, fast mode is 1.5 times the speed. GPT-5.6 and GPT-5.5 then consume credits at 2.5 times the standard rate, and GPT-5.4 at twice.
| Situation | Worth it? |
|---|---|
| You are sitting there waiting, mid-debug | Yes. You are the expensive part |
| A long unattended run | No. Nobody is waiting, and the bill is the only thing that changes |
| A demo in front of people | Probably |
Reasoning effort
Higher effort can improve results on hard work, and it takes longer and uses more tokens. Start at the default and raise it for the task that needs planning, rather than raising it globally and forgetting.
What actually reduces the bill
- Say where to look. A prompt that names the file skips the searching, from lesson 8.
- Keep
AGENTS.mdshort. It is read at the start of every session, from lesson 6. - Let skills stay unloaded. Only names and descriptions cost you until one is used, from lesson 14.
- Watch the cached input. The usage in lesson 9 shows it, and a CI job that reruns on every push pays far less when the prefix is cached.
Access is per account
A model your account cannot use fails at the first call, not at sign-in. Lesson 0 showed exactly that error, and it is worth recognising before you spend an hour on your config.
Try it yourself
- Run the same task at default effort and one step higher. Compare the result and the time.
- Check where your Codex usage is counted before a heavy week, not after it.
This is what real progress feels like.