OpenAI Launches GPT-5-Codex for Smarter Coding
OpenAI has introduced GPT-5 Codex, a new version of its AI coding agent designed to handle coding tasks more effectively. Unlike earlier models, this version dynamically adjusts its “thinking” time, ranging from just a few seconds to as long as seven hours. Consequently, it delivers stronger results on advanced coding benchmarks, especially in agentic tasks.
Broader Availability Across Platforms
The rollout of GPT-5-Codex is already underway in Codex products, including terminals, IDEs, GitHub, and ChatGPT. Users with Plus, Pro, Business, Edu, and Enterprise plans can access it now. Plans are in place to expand access for API customers, making the model widely available in the future. This move also positions Codex to compete more directly in the rapidly growing AI coding tool market, which has seen rising adoption and major investments over the past year.
Stronger Benchmark Performance
According to recent evaluations, GPT-5-Codex outperforms GPT-5 on benchmarks such as SWE-bench Verified, which assesses agentic coding skills. It also excels in large-scale code refactoring tasks, showing notable improvements in both speed and accuracy. Additionally, the model has been trained to handle code reviews, where assessments indicate it produces fewer incorrect comments and more impactful feedback. By adapting its workload in real time, GPT-5-Codex optimizes computational resources and boosts productivity for developers.
