โ† sources

๐Ÿ”ฅ GPT keeps quietly shrinking your limits. Meanwhile, the neighbors are stretching theirs.

source: reddit (reddit:OpenAI)published 2026-11-14 05:20ingested 2026-11-15 02:35

retained text

Claude Code's 5-hour limit might stretch further this time.

v2.1.323 adds an experimental feature: Code Mode.

Before: one tool call per turn โ€” read the result, then decide what's next.

Now: one chunk of JS runs several calls in a single pass, in parallel.

Here's the key part:

Anthropic's own docs say it โ€” every turn resends the whole conversation so far.

The bigger the context, the more tokens you burn.

Fewer round trips, fewer resends.

And a big chunk of useless output never even enters the context.

The limit itself hasn't changed, but the same limit now gets you a lot more done.

For users, that's a limit raise in everything but name. ๐Ÿ’ช

One catch nobody mentions:

It's off by default, and only works in interactive terminal sessions.

One line to turn it on:

CLAUDE\_CODE\_EXPERIMENTAL\_CODE\_MODE=1 claude

Official docs: https://code.claude.com/docs/en/code-mode

document fetched 2026-11-15 00:26 ยท canonical https://reddit.com/r/OpenAI/comments/1xvu5pt/gpt_keeps_quietly_shrinking_your_limits_meanwhile