OpenAI Reports 56 Percent Token Efficiency Gain for GPT-5.5 in Perplexity Workflows

OpenAIOpenAI

· Updated

Perplexity built an internal tool in under an hour using GPT-5.5 within the Codex platform. The model completed complex computer-use tasks with 56% fewer tokens, significantly reducing latency and improving feedback loops for end users.

OpenAI shared performance data from Perplexity, an AI-powered search engine, demonstrating the efficiency of the new GPT-5.5 model in agentic workflows. This efficiency mirrors its role as a default orchestrator for Perplexity's Computer platform, allowing the team to build an internal tool in under an hour using Codex, which extends agentic coding.
Token efficiency
56% fewer tokens
Internal tool build time
Under 1 hour
Primary capability
Computer Use
Model
GPT-5.5
Access
Codex API and desktop app

The update highlights a 56% reduction in token usage for "Perplexity Computer" workflows—tasks involving computer use, building on the model's ability to control desktop interfaces. This efficiency is critical as agentic loops often suffer from high latency. By using fewer tokens, GPT-5.5 creates faster feedback loops while completing the same complex tasks.

You can leverage these efficiency gains by accessing GPT-5.5 through the Codex API or desktop application. The model is optimized for technical builds where reducing token volume translates to faster execution. Access is currently available for paid users on ChatGPT and Codex plans.

View the full update on openai.com
OpenAI Developers
OpenAI Developers
@OpenAIDevs
X

At @perplexity_ai, GPT-5.5 in Codex helped build an internal tool in under an hour. In Perplexity Computer workflows, GPT-5.5 used 56% fewer tokens on the same complex tasks, creating faster feedback loops for users. https://t.co/iEuZ9ttsRo

8retweets192likes
View on X

Still wondering? A few quick answers below.

GPT-5.5 is an OpenAI model optimized for agentic tasks and computer use within the Codex platform. It is designed to autonomously plan and execute multi-step actions, such as writing code or navigating digital environments. The model focuses on efficiency, allowing it to complete complex technical tasks with significantly fewer tokens than previous versions.

In real-world workflows at Perplexity, GPT-5.5 demonstrated a 56 percent reduction in token usage for the same complex tasks. This improvement in token efficiency directly reduces the amount of data processed during agentic loops. This reduction results in faster feedback loops for users and more streamlined execution of multi-step computer workflows.

Developers can use GPT-5.5 within Codex to build internal tools and automate engineering workflows. For example, Perplexity used the model to develop a new internal tool in less than one hour. The model is particularly effective for computer-use tasks, where an AI agent must interact with software interfaces to achieve a goal.

The 56 percent reduction in token usage primarily benefits users by creating faster feedback loops. Because the model requires fewer tokens to complete complex tasks, the time spent in the agentic loop is reduced. This leads to quicker responses and more efficient interactions when using AI agents for multi-step computer and coding workflows.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

OpenAI

See all AI news & updates from OpenAI →

Share this update