OpenAI: Our new model GPT-5-Codex-Mini – a more cost-efficient GPT-5-Codex (github.com)

OpenAI announced GPT-5-Codex-Mini, a compact, more cost-efficient variant of GPT-5-Codex aimed at lowering compute and inference costs for code generation and developer-assistant use cases. The release is positioned as a drop-in lighter-weight option that preserves Codex capabilities while reducing resource requirements, making it attractive for higher-volume, latency-sensitive, or cost-constrained deployments (e.g., CI automation, developer tools, and edge-like environments). The rollout is accompanied by a flurry of platform and SDK changes that make integration and tuning easier: app-server v2 APIs (Thread/Turn and account/login endpoints), improved session metadata and token refresh handling, and TypeScript SDK support for a modelReasoningEffort option plus support for models with a single reasoning effort. Other updates include a “model nudge” feature for query shaping, sandbox and build fixes, TUI/CLI usability improvements, and documentation/contributing clarifications (including a note that gpt-5-codex won’t amend commits unless requested). Collectively, these changes signal readiness of both the model and the platform for production use — enabling developers to trade off cost, latency, and reasoning effort more granularly when deploying Codex-style assistants.