Skip to content
Trending storyObserving

GPT 6.1 的 token 效率秘密：靠上下文剩余量施压？

1 report1 reporting sourceUpdated 1 day ago

Understand the story

Report excerpt

有用户观察到 GPT 6.1 似乎更省 token，猜测原因是后训练或注入机制让模型感知到上下文剩余量，从而产生压力。例如模型会主动提及"还剩约 26k""37000 token 上下文已用 28000、剩 18000"等剩余额度，并据此压缩工作范围。

From Reddit · ClaudeCode / Codex / VibeCoding

Report timeline

Follow the reports to explore different perspectives.

10/5
  1. Reddit · ClaudeCode / Codex / VibeCoding
    GPT 6.1 的 token 效率秘密：靠上下文剩余量施压？

    有用户观察到 GPT 6.1 似乎更省 token，猜测原因是后训练或注入机制让模型感知到上下文剩余量，从而产生压力。例如模型会主动提及"还剩约 26k""37000 token 上下文已用 28000、剩 18000"等剩余额度，并据此压缩工作范围。

Interest in this story

Current interest 5·Peak within comparable coverage 6（Oct 6, 16:00）·Change over 24 hours within comparable coverage –

024610/616:0010/619:0010/621:0010/700:00

The trend compares the same participants under continuous, complete observation, so its coverage may be narrower than the current score. Hover or tap to view hourly interest; use the left and right arrow keys to navigate.