GPT-5.5 Is OpenAI's Strongest Agentic Coding Model - Codex Gets the Upgrade Today

OpenAI has released GPT-5.5, positioning it as the company's most capable agentic coding model to date. The model demonstrates significant improvements in automated programming tasks, scoring 82.7% on Terminal-Bench 2.0, a benchmark specifically designed to evaluate performance on long-running command-line operations. The new model represents a clear evolution from GPT-5.4, with enhanced end-to-end coding capabilities integrated into OpenAI's Codex platform. GPT-5.5 can autonomously read existing codebases, implement necessary changes, debug issues, run tests, and validate results within a single operational cycle. This streamlined approach reduces the need for human intervention across the development pipeline. On SWE-Bench Pro, which measures real-world GitHub issue resolution capabilities, GPT-5.5 achieved a 58.6% success rate. While this represents a modest improvement over GPT-5.4's 57.7% score, OpenAI claims the new model reaches this performance level whilst consuming fewer tokens, indicating improved efficiency. Perhaps most notably, OpenAI asserts that GPT-5.5 maintains equivalent per-token latency to its predecessor in production environments. This claim challenges the typical trade-off between model capability and response speed, suggesting meaningful optimisations in the underlying architecture. The development signals continued progress in the competitive landscape of AI-powered coding tools, where the focus remains on extending autonomous operation periods without sacrificing performance.