Perplexity has quietly rolled out support for OpenAI‘s latest models — GPT-5.4 and GPT-5.4 Thinking — for its Pro and Max subscribers, giving users on the platform access to what is arguably one of OpenAI’s most capable releases to date.

OpenAI officially launched GPT-5.4 on March 5, 2026, calling it the third major update to the GPT-5 foundational model. What makes this one stand out from the rest is that it is the first general-purpose model from OpenAI to come with native computer-use capabilities — meaning it can actually operate software, issue keyboard and mouse commands, and interact with applications just by analyzing screenshots. On the OSWorld-Verified benchmark, which tests exactly this kind of desktop navigation ability, the model scored a 75 percent success rate.

But it does not stop there. GPT-5.4 also brings improved reasoning, stronger coding chops (inherited from GPT-5.3 Codex), and better performance across spreadsheets, documents, and multi-tool workflows. It supports a massive 1 million token context window, making it especially useful for long-horizon tasks that require a lot of information to be kept in mind at once. There is also a new “tool search” feature that dynamically pulls in only the tools the model needs at a given moment, reducing token usage by as much as 47 percent in large tool ecosystems.

GPT-5.4 Thinking, the reasoning-focused sibling, is where things get even more interesting for power users. It introduces a concept called steerability — before diving into a complex query, the model first lays out its thinking plan upfront, and users can actually jump in mid-response to redirect or adjust its course. That is a meaningful shift from how AI reasoning has typically worked, where you would have to wait for the full output and then start over if something went off track. The model also has significantly improved deep web research, better context retention during long tasks, and two adjustable modes — Standard for everyday speed-accuracy balance, and Extended for when you really need the model to think hard before responding.

For Perplexity subscribers specifically, GPT-5.4 joins an already packed model lineup that includes Sonar, Gemini 3.1 Pro, Claude Sonnet 4.6, Claude Opus 4.6, and Kimi K2.5, among others. Having GPT-5.4 in that mix means Pro and Max users now have access to OpenAI’s frontier model right from within the Perplexity interface, without needing a separate ChatGPT subscription. That is a solid value add, especially given that GPT-5.4 Thinking in ChatGPT is otherwise reserved for Plus, Team, Pro, and Enterprise tiers, and GPT-5.4 Pro sits even higher — exclusive to ChatGPT Pro and Enterprise users.

It is worth noting that OpenAI classified GPT-5.4 as “High cyber capability” and has deployed it with an expanded safety stack, including monitoring systems, trusted access controls, and asynchronous blocking for high-risk requests. The company also introduced a new open-source evaluation called chain-of-thought controllability to help ensure the model’s reasoning remains transparent and cannot be deliberately obfuscated.​

For anyone already on Perplexity Pro or Max, the update requires no extra steps — GPT-5.4 and GPT-5.4 Thinking simply appear as selectable options in the model picker.