There has been a lot of brouhaha in e.g. Reddit about todays (10.4.2026) changes to plans (OpenAI announced new 100€ ‘pro lite’ tier, and there was worry that ‘plus’ tier was cannibalized for it). So I decided to use 90% of my 5 hour quota using fast mode gpt-5.4 (high for planning, medium for implementation).

How much did I get out of it?

┌──────────────┬──────────────────────────────────────────────────────┬──────────────┬────────────┬────────────┬────────────────┬────────────────┬─────────────┐
│ Date         │ Models                                               │        Input │     Output │  Reasoning │     Cache Read │   Total Tokens │  Cost (USD) │
├──────────────┼──────────────────────────────────────────────────────┼──────────────┼────────────┼────────────┼────────────────┼────────────────┼─────────────┤
│ Apr 10, 2026 │ - gpt-5.4                                            │    1,069,508 │     94,045 │     38,453 │     13,131,648 │     14,295,201 │       $7.37 │
└──────────────┴──────────────────────────────────────────────────────┴──────────────┴────────────┴────────────┴────────────────┴────────────────┴─────────────┘

Quite a lot. This was 91% of my 5 hour quota, and 14% of weekly limit, and it is already 7$ in API token costs. I’m not sure if the ccusage tool accounts for ‘fast’ mode which I used, but even if it does, this is quite decent showing. 14 million (mostly cached tokens) for (roughly) 2,8% of my monthly quota, which means that this essentially cost me perhaps 0,7€ or so.

Thoughts

Either

  • ccusage doesn’t account for fast mode,
  • this sample is too small, or
  • the amount of tokens you get is perhaps halved for the same quota (based on the previous monthly analysis of late February / most of March).

Still, given typical 1:10 noncached:cached pricing structure, paying 23€ monthly for this is still crazily good deal as of now. Especially when I am not actively developing all the time, fast mode Codex seems to be still the best deal in town (I am sort of missing Spark, which I no longer have access to - it was very nice for fast and simple things, even if it got things wrong often).