OpenAI Unveils GPT-5.4 With 1M Token Context and ZRD surface

The OpenAI team has upgraded their latest GPT model to GPT-5.4, introducing a more capable, highly efficient architecture specifically designed for developers. This release is being hailed as OpenAI's most streamlined and efficient model to date, striking a perfect balance between raw reasoning power and operational speed.

Building on the legacy of the GPT-3.3 Codex, the GPT-5.4 update represents a monumental shift in how AI interacts with multi-layered software environments and professional tools.

This version moves beyond simple text generation to offer seamless integration across complex platforms, enabling it to execute sophisticated tasks within spreadsheets, dynamic presentations, and technical documentation with unprecedented precision.

Central to this evolution is a breakthrough in token differentiation, enabling the model to distinguish between "Logic Tokens" and "Content Tokens" and navigate diverse digital workflows with a massive 1-million-token context window.

This high-density architecture enables the model to ingest entire codebases or massive datasets with fewer resources, resulting in significant cost savings and near-instant response times, effectively transforming it from a passive assistant into a highly efficient autonomous agent.

Starting from March 5 2026, GPT-5.4 Thinking is officially available to ChatGPT Plus, Team, and Pro users, effectively replacing GPT-5.2 Thinking as the flagship reasoning model.

While the more powerful GPT-5.4 Pro is reserved for Pro and Enterprise tiers, the standard 5.4 model remains accessible for all paid subscribers, with the legacy GPT-5.2 version scheduled for retirement on June 5, 2026.

While these advancements represent a significant leap in logic and efficiency, the context windows for this rollout remain consistent with the previous generation, ensuring a stable and familiar experience for power users as they transition to this more capable architecture.

Regarding pricing, GPT-5.4 follows a token-based usage model. At the same time, the cost per token is slightly higher than GPT-5.2, but it delivers significantly superior accuracy and a higher return on investment for complex tasks.

OpenAI has announced that GPT-5.4 includes experimental support for a 1-million-token context window, allowing developers to push the boundaries of large-scale data processing.

To utilise this feature, developers can begin testing by configuring the model_context_window and model_auto_compact_token_limit parameters within their API settings.

This release includes an expanded cyber-safety stack, featuring advanced monitoring systems, trusted access controls, and asynchronous blocking for high-risk requests. These protections are specifically optimised for customers on the Zero Data Retention (ZDR) surface, ensuring that privacy and security remain uncompromised.

Alongside ongoing investments in their broader security infrastructure, the OpenAI team has provided formal assurances regarding the safety and integrity of the model's outputs.

GPT-5.4 sets a new industry benchmark by prioritising security-rich architecture and operational efficiency. It provides developers with the most effective tools to date while ensuring that safety and data integrity remain at the core of every interaction.