OpenAI Unveils GPT-5.4, a Stronger Model for Reasoning, Coding, and Real Work

TL;DR

OpenAI released GPT-5.4, its new flagship model for ChatGPT, the API, and Codex.

It improves reasoning, coding, and agent workflows, introduces native computer-use capabilities, supports up to 1M tokens of context, and adds tool search to make large tool ecosystems more efficient.

The model is more accurate, more token-efficient, and better at real professional tasks like coding, spreadsheets, documents, and web research. A higher-performance GPT-5.4 Pro version is also available for complex workloads.

Key Points

Highlight key points with color coding based on sentiment (positive, neutral, negative).

GPT-5.4 introduces advanced computer-use capabilities, allowing developers to build agents that can operate computers and execute complex workflows across applications, enhancing productivity in professional environments.

The model supports up to 1 million tokens of context, enabling it to handle long and complex tasks more efficiently, which is beneficial for developers working on projects that require extensive context maintenance.

GPT-5.4 is more token-efficient than its predecessors, using fewer tokens to solve problems, which translates to faster processing speeds and reduced costs for developers using the API.

The model includes improved tool search and tool calling features, allowing it to work more effectively with a wide range of external tools and complete multi-step workflows with lower latency and cost.

GPT-5.4 excels in coding tasks, particularly in frontend development, providing more aesthetic and functional results, and offers faster token velocity in Codex, which helps developers maintain workflow efficiency during coding and debugging.

OpenAI has released GPT-5.4, a new iteration of its language model, now accessible through ChatGPT, the API, and Codex. This model supports up to 1 million tokens of context, allowing it to manage complex task planning and execution over extended periods. It is designed to be more token-efficient, which results in faster and higher-quality outputs.

GPT-5.4 shows improved capabilities in deep web research, particularly for queries requiring detailed exploration. It maintains context more effectively, which benefits tasks that involve extended reasoning. The model's computer-use capabilities have been improved, allowing agents to manage complex workflows across various applications. Its tool search feature helps agents locate and use the appropriate tools more efficiently.

Another important improvement concerns how GPT-5.4 works with large tool ecosystems such as Model Context Protocol (MCP) servers. MCP is an open standard that allows AI models to interact with tools, APIs, and data sources through a unified interface. With GPT-5.4’s tool search mechanism, the model no longer needs to load every tool definition into the prompt. Instead, it can dynamically discover and retrieve the tool specifications it needs at runtime. This approach reduces token usage and improves performance when working with environments that may expose dozens or even hundreds of tools through MCP.

In terms of performance, GPT-5.4 surpasses its predecessors, achieving higher scores in benchmarks such as GDPval, SWE-Bench Pro, and OSWorld-Verified. These scores highlight its proficiency in professional knowledge work, coding, and computer-use tasks. The model is also more factual, reducing the likelihood of generating false claims compared to earlier versions.

For coding tasks, GPT-5.4 integrates the strengths of GPT-5.3-Codex with improved knowledge work and computer-use functionalities. It performs well in complex frontend tasks, producing results that are both aesthetic and functional. The model's tool use is more precise, with improved tool calling and the ability to complete multi-step workflows with reduced cost and latency.

GPT-5.4 is available in various configurations, including GPT-5.4 Pro, and is gradually being rolled out across ChatGPT and Codex. Pricing reflects its advanced capabilities, and the model is designed to be steerable, allowing users to adjust its direction mid-response.

Key Numbers

Present key numerics and statistics in a minimalist format.

1,000,000 tokens

Maximum context window supported by GPT-5.4.

83.0 percent

GPT-5.4 score on GDPval benchmark measuring professional knowledge work.

70.9 percent

GPT-5.2 score on GDPval benchmark for comparison.

57.7 percent

GPT-5.4 score on SWE-Bench Pro (public) coding benchmark.

75.0 percent

GPT-5.4 score on OSWorld-Verified benchmark measuring computer-use abilities.

72.4 percent

Human performance on OSWorld-Verified benchmark.

82.7 percent

GPT-5.4 score on BrowseComp benchmark measuring agentic web browsing ability.

89.3 percent

GPT-5.4 Pro score on BrowseComp benchmark.

33 percent

Reduction in false individual claims compared to GPT-5.2.

18 percent

Reduction in responses containing factual errors compared to GPT-5.2.

47 percent

Token reduction achieved when using tool search with MCP servers.

93.7 percent

GPT-5.4 score on ARC-AGI-1 benchmark for abstract reasoning.

73.3 percent

GPT-5.4 score on ARC-AGI-2 benchmark for abstract reasoning.

92.8 percent

GPT-5.4 score on GPQA Diamond benchmark.

2.50 USD / 1M token

Input token price for GPT-5.4 in the API.

15 USD / 1M token

Output token price for GPT-5.4 in the API.

30 USD / 1M token

Input token price for GPT-5.4 Pro in the API.

180 USD / 1M token

Output token price for GPT-5.4 Pro in the API.

Stakeholder Relationships

An interactive diagram mapping entities directly or indirectly involved in this news. Drag nodes to rearrange them and see relationship details.

Organizations

Key entities and stakeholders, categorized for clarity: people, organizations, tools, events, regulatory bodies, and industries.

OpenAI AI Research Organization

OpenAI is the company behind the development and release of GPT-5.4, a new frontier AI model designed for professional work.

Tools

Key entities and stakeholders, categorized for clarity: people, organizations, tools, events, regulatory bodies, and industries.

GPT-5.4 AI Model

GPT-5.4 is OpenAI’s latest frontier model combining advanced reasoning, coding capabilities, agent workflows, and native computer-use features.

ChatGPT AI Application

ChatGPT is OpenAI’s conversational AI application where users can interact with models like GPT-5.4 for knowledge work, coding, research, and productivity tasks.

Codex AI Coding Environment

Codex is OpenAI’s coding environment and agent system that uses models like GPT-5.4 to assist with software development and automate coding workflows.