openai / openai/codex

5.6 Terra High has become stupid and consumes all tokens on simple tasks

Open
#40,790 1 comment 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

app bug model-behavior rate-limits
Dominant language
Rust
Stars
125k
Forks
19.4k
PR merge metrics
PR metrics pending

Description

What issue are you seeing?

5.6 Terra High has become stupid and consumes all tokens on simple tasks.

Having used the 5.6 Terra model at length doing similar tasks, I've noticed an enormous change in its behaviour very recently.
It now churns through tokens like they're made of air and gets nowhere.

Just now I tasked it to compile 15 sentences into a docx. following a protocol it has followed all month with no problems.
It didn't follow the protocol, made up its own, erroneously printed each on a separate page instead of all on one page / with non-wasted space.

On pointing out its failure and requesting it sticks to ordinary workflow, it acknowledged and then proceeded to use 5% (from 9% to 4%) of 5h usage and created a newly exported version with two per page. still completely incorrect on a simple task of pagination.
On reiterating the simple task, it consumed the remaining 4% of 5h usage and ceased working.

What steps can reproduce the bug?

Request any simple task be done using Terra 5.6 on Codex Desktop for Mac

What is the expected behavior?

Continuing to follow simple instructions, and not burn tokens doing nothing.

Additional information

No response

Contributor guide

Open the contributing guide

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

Start by reproducing the issue in Codex Desktop for Mac with the Terra 5.6 model using a simple task that follows the reported protocol. Done means the task follows the instructions, produces the expected pagination, and does not consume excessive tokens.

Written by the indexing model from the issue text.

Assessment

Tech stack
macos
Domain
ai, desktop
Issue type
Bug
Difficulty
5/5
Estimated time
Over a week
Activity status
Active
Clarity
Needs clarification
Newbie friendliness
25/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.