This September, OpenAI released GPT-5-Codex, a version of the GPT-5 architecture specially optimized for the “autonomous coding” tasks on the Codex platform. The model is designed for real-world software engineering workflows — from initiating new projects and adding features/tests, to performing large-scale code refactors.
Now, OpenAI is rolling out GPT-5-Codex-Mini, a smaller, lower‐cost variant of GPT-5-Codex. As the name suggests, the Mini version sacrifices a modest amount of performance in exchange for roughly 4× higher usage allowance for developers.
In benchmark comparisons on the SWE-bench Verified test:
- GPT-5 High scored ~ 72.8%
- GPT-5 Codex scored ~ 74.5%
- GPT-5 Codex-Mini scored ~ 71.3% (per the leaked figures)
OpenAI is positioning Codex-Mini for lighter-weight engineering tasks or for situations where usage is nearing rate limits. When usage of the Mini version hits about 90% of capacity, the system will automatically prompt users to switch to the full Codex model.
The Mini edition is already available in CLI and IDE extensions; API support is coming soon. Meanwhile, thanks to improved GPU efficiency, ChatGPT Plus, Business, and Edu users will enjoy a ~50% increase in rate-limits; ChatGPT Pro and Enterprise users receive priority scheduling for even lower latency.
Additionally, OpenAI has upgraded the backend of Codex to ensure consistent, predictable availability, addressing previous issues around cache misses and routing fluctuations.
The above content is compiled by ModeZone, a fashion and entertainment magazine.
