Codex Repeatedly Compresses and Burns Tokens in Long Tasks
aigclink · x · 2026-07-19
A post pointed out that Codex can insanely burn through tokens in long tasks due to a flaw in its compression design: after compression, the model only sees the initial instructions, causing it to repeatedly enter an "initial instructions -> compress -> execute initial instructions" loop. It might only complete the task by chance if it skips compression; otherwise, it drains the entire quota.
The accompanying image further illustrates this runaway loop: once the context is compressed, the working memory is repeatedly reset, leading to redundant execution and uncontrolled token usage. The poster tagged relevant personnel, hoping they would look into the issue.
Related event: Alibaba Announces 2.4T Open-Weight Model Qwen3.8(22 posts)→
More from coding & agent
- 'My intelligence stack': developer splits AI coding work across Opus 5.5, Fable and Astra with human review gates — sh_reya · 2026-10-27
- Baseten webinar: evaluate open models for coding agents by benchmarking on your own merged PRs — baseten · 2026-10-27
- Thorsten Ball on working with agents: run more evals, automate everything, use the best models — omarsar0 · 2026-10-27
- Where to draw the approval line for scheduled agent jobs — Reddit discussion — MemStrata · 2026-10-27
- 4 MCP servers worth a look: full-text YouTube search, Chrome session reuse, Framer editing — Prior_Engineering486 · 2026-10-27
- DeepMind paper: LLMs can't sample randomly on their own — pass random numbers from code instead — rohanpaul_ai · 2026-10-27