CEO of OpenAI, Sam Altman, officially announced the launch of a new programming large model GPT-5.3-Codex. As the latest iteration of the Codex series, this model has achieved breakthroughs in both technical indicators and practical applications, marking a more mature stage in AI's role in assisting programming and computer automation.

image.png

In terms of performance, GPT-5.3-Codex has set new records in multiple authoritative evaluation benchmarks. It achieved a score of 57% in the SWE-Bench Pro programming test, and in the TerminalBench2.0 and OSWorld benchmarks, which focus more on system operations, it scored 76% and 64%, respectively. This means that the model not only writes code but also has excellent computer operation capabilities, allowing it to understand and execute complex operating system tasks like human engineers.

Notably, the new model shows remarkable efficiency. OpenAI stated that when completing tasks of the same complexity, the token consumption of GPT-5.3-Codex is reduced by more than half compared to the previous 5.2 version, while the processing speed per token has increased by over 25%. This "high-speed and low-energy" feature will significantly reduce the cost for enterprises and developers integrating AI programming capabilities. Additionally, the model supports real-time control and dynamic updates during task execution, greatly enhancing the flexibility of the development process.

In terms of security, GPT-5.3-Codex is the first model from OpenAI to be rated "high-level" in the cybersecurity dimension of the security protection framework. To further build a defense system, OpenAI has also launched a pilot program for a trusted access framework and allocated $10 million in API credits, aiming to accelerate global cybersecurity defense construction through AI technology.

Key points:

  • 💻 Double Evolution in Programming and Practical Operations: The model has set new records in multiple evaluations such as SWE-Bench Pro, demonstrating mature autonomous operation capabilities in computer systems and complex programming skills.

  • ⚡ Dramatic Improvement in Operational Efficiency: Compared to the 5.2 version, the token consumption for the same task is reduced by more than 50%, and the processing speed is improved by over 25%, significantly enhancing the cost-effectiveness of task execution.

  • 🛡️ Top-Level Security Defense: The first model to receive a "high-level" rating in the cybersecurity dimension, OpenAI has allocated $10 million in credits to support the construction of cybersecurity systems.