OpenAI Optimizes ChatGPT and Codex Desktop App Disk Storage by 60%
OpenAI team member Tibo announced that optimizations have been launched for the disk state storage of the ChatGPT and Codex desktop applications, expected to reduce storage by approximately 60%.
This update is part of OpenAI's 28-day product sprint on day 9. Tibo referred to it as a small-scale release but noted that sufficient feedback had been received regarding heavy users running out of disk space. The optimization targets the way the application stores session states on local disks.
Previous user reports indicated that Codex session files could grow to over 60GB, with some users manually cleaning up to reduce usage from over 60GB to 30GB; other users reported archiving about 200GB of uncompressed Codex chat logs to NAS. Reports on the OpenAI community forum also indicated that after the update, daily session folders could reach tens to hundreds of GB, with individual jsonl files ranging from 750MB to 1.65GB.
The optimization has taken effect for the desktop application without requiring additional user settings. Tibo's post received over 350,000 views within hours of publication. Some users in the replies indicated that this freed up space for long coding sessions, while others felt it was more of a fix for existing issues rather than a new feature.
The buyers are developers and heavy users relying on local desktop applications for long sessions; the seller is OpenAI, which provides the ChatGPT and Codex desktop clients. The resource flow shifts from high-occupancy session states on local disks to a more compact storage method, benefiting users with limited local disk space, while those still relying on cloud project history or complete session logs face pressure.
Previous sprints also included an increase in the Sol context window from 272k to 500k on day 7, as well as multiple usage limit resets.
Source: Public Information
ABAB AI Insight
OpenAI is continuously iterating to expand ChatGPT and Codex from web conversations to desktop applications, supporting local state persistence to maintain long sessions and sub-agent workflows. Tibo, as a member of the Codex and ChatGPT related teams, publicly responded to disk usage feedback during the 28-day sprint, where reports of daily session folders exceeding 100GB and individual files over 1GB had already emerged in the community.
In terms of capital pathways, OpenAI is investing engineering resources into local storage optimization rather than solely enhancing model capabilities to reduce hardware friction for heavy users. The motivation is that the desktop application has become the entry point for developers' workflows, and running out of disk space can interrupt long coding or agent tasks; by compressing session state storage, it reduces the cost for users of manual cleanup or log migration. Resources are extending from cloud quota management to local client storage efficiency.
This is analogous to other locally prioritized development tools optimizing caching and logging after user growth, or browsers introducing storage cleanup mechanisms after historical data inflation. Currently, OpenAI is transitioning AI tools from lightweight conversations to heavy local execution, with storage becoming a new constraint.
Essentially, this is a supportive adjustment of technical substitution: as model and agent capabilities improve, local state storage becomes a bottleneck, and optimization shifts value from raw data accumulation to scalable session persistence. The mechanism is that when session files grow linearly due to agent proxies and historical accumulation, storage efficiency directly determines whether users can continue using the service; a 60% reduction frees up local space, but cloud history and project-level storage still pose independent constraints.
ABAB News · Cognitive Law
After model capabilities improve, local storage becomes the bottleneck
Small-scale optimizations can resolve hardware friction for heavy users
Sprint-style releases directly translate user feedback into storage efficiency