問題修復
Codex 複製的對話與思考字數不再被重複算錢
你用 Codex 分支（fork，把先前的對話複製一份出來繼續聊）時，系統不會再把舊的歷史對話當成新用量多算一次錢了；同時，模型在背後思考時用的字數（reasoning token），也不會再被重複計價。以前系統會笨笨地把複製過來的舊內容跟已經算過錢的思考字數再算一遍，現在它懂得剔除這些水分，只算真正新增的花費。就像你拿舊的草稿影印來寫新文章，計費員以前連草稿上的字都跟你收錢，現在只跟你收新寫上去的字數錢。
繁體中文
English
BUGFIX
Forked Codex chats and reasoning tokens are no longer double-charged
When you use Codex to fork a conversation (fork — making a copy of an existing chat to take it in a new direction), the system no longer double-charges you for the old history it pulled over; additionally, the tokens the model uses to think behind the scenes (reasoning tokens) are no longer billed twice. Previously, the system would mindlessly charge you for the copied legacy content and the already-priced reasoning words, but it now strips out that fluff and bills only for truly new expenditure. It's like photocopying an old draft to start a new essay: the clerk used to charge you for the words already printed on the draft, but now you only pay for the fresh ink you add.
原始 CHANGELOG
Original changelogCodex 複製的對話與思考字數不再被重複算錢
**Forked Codex conversations no longer replay parent history as new usage**: Codex can embed a timestamp-rewritten copy of the parent conversation in a fork JSONL. The loader now matches and excludes that replay while retaining both the original parent usage and new post-fork usage. (#40, by @ericweichun)
**Codex reasoning tokens are no longer charged twice**: `reasoning_output_tokens` is already included in Codex's `output_tokens`, so JSONL and SQLite usage readers now price the output total once. (#40, by @ericweichun)