OpenAI has released GPT-5.4 Thinking, the latest reasoning model in its GPT-5 series. This model represents a significant advancement in safety features, becoming the first general-purpose model to implement mitigations specifically for high capability in cybersecurity threats. The comprehensive safety approach builds upon previous GPT-5 iterations while adding new protections.

According to OpenAI's system card documentation, GPT-5.4 Thinking extends the safety framework established with GPT-5.3 Codex and continues the evolution seen in ChatGPT and the API. The model's cybersecurity mitigations address sophisticated threats that could emerge from highly capable AI systems. These safeguards aim to prevent potential misuse while maintaining the model's reasoning capabilities.

OpenAI notes that there is no GPT-5.3 Thinking model, making GPT-5.2 Thinking the primary baseline for comparison. The company has published detailed system cards explaining the safety approach for this model, similar to previous releases in the series. This transparency allows researchers and users to understand the specific protections implemented for GPT-5.4 Thinking's enhanced capabilities.