OpenAI has released GPT-5.4 Thinking, a frontier reasoning model that combines advanced coding capabilities with improved research and planning features, now available in ChatGPT with the ability to show users its thinking process upfront so they can adjust course mid-response. The model represents a significant step forward in how reasoning models handle complex, real-world tasks like coding, spreadsheet work, and professional document creation.

What Makes GPT-5.4 Thinking Different From Previous Reasoning Models?

GPT-5.4 Thinking brings together several major improvements that set it apart from earlier versions . The model incorporates the industry-leading coding capabilities of GPT-5.3-Codex, a specialized coding agent released in February 2026, while enhancing how it works across tools, software environments, and professional tasks. This means the model can now handle spreadsheets, presentations, and documents with greater accuracy and efficiency than before.

One of the most notable features is the upfront planning capability. In ChatGPT, GPT-5.4 Thinking can now provide a plan of its thinking process before it works through a problem, allowing users to adjust course mid-response while the model is still working. This addresses a long-standing frustration with AI reasoning models: users often don't see where the model is headed until it's too late to correct course, requiring additional back-and-forth exchanges to get the desired result.

How Does GPT-5.4 Thinking Improve on Web Research and Context Management?

The model delivers several practical improvements for everyday use . GPT-5.4 Thinking improves deep web research, particularly for highly specific queries that require nuanced information gathering. It also better maintains context for questions that require longer thinking processes, which is critical when tackling multi-step problems or research tasks that span multiple topics.

Context window management has been enhanced to support longer thinking sessions. This means the model can process and reason through larger amounts of information without losing track of earlier details in the conversation. The result is higher-quality answers that arrive faster and stay relevant to the task at hand, reducing the need for clarification or follow-up questions.

How to Maximize GPT-5.4 Thinking for Your Workflow

• Use the Thinking Preview Feature: Take advantage of the upfront planning capability by reviewing the model's proposed approach before it completes the response. This allows you to provide feedback or redirect the reasoning early, saving time on revisions.
• Leverage for Complex Professional Tasks: Deploy GPT-5.4 Thinking for spreadsheet analysis, presentation creation, and document editing where accuracy and efficiency matter most. The model's improved coding and tool integration makes it particularly strong for these workflows.
• Optimize for Specific Web Research: When you need deep research on niche topics or highly specific queries, GPT-5.4 Thinking's improved research capabilities can deliver more relevant results than general-purpose models, reducing the need for manual fact-checking.
• Choose Thinking Levels Based on Task Complexity: Users can select different thinking time settings (Light, Standard, or Extended) depending on whether they need faster responses or more thorough reasoning. Match the setting to your task's complexity requirements.

What's Happening With Older OpenAI Models?

OpenAI has been retiring older models to streamline its product lineup . As of March 11, 2026, GPT-5.1 models are no longer available in ChatGPT, including GPT-5.1 Instant, GPT-5.1 Thinking, and GPT-5.1 Pro. Existing conversations that used these models automatically continue on corresponding current models: GPT-5.3 Instant, GPT-5.4 Thinking, or GPT-5.4 Pro.

Additionally, GPT-4o and other legacy models were retired on February 13, 2026. This includes GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini from ChatGPT. The company also retired GPT-5 (both Instant and Thinking versions). These changes do not affect the API at this time, so developers using these models through the API should not experience disruptions.

How Is OpenAI Addressing User Experience Beyond Raw Performance?

Beyond benchmark improvements, OpenAI has been making subtle but important updates to how its models communicate . GPT-5.3 Instant received updates in March 2026 that improve follow-up tone and reduce teaser-style phrasing in responses. The model now avoids phrases like "If you want...," "You'll never believe...," or "I can tell you these three things that..." which can interrupt conversational flow.

Earlier updates to GPT-5.2 Instant focused on response style and quality, delivering more measured and grounded tones that are contextually appropriate to conversations. The model also improved at providing clearer, more relevant answers to advice-seeking and how-to questions, placing the most important information upfront. These changes reflect user feedback on aspects of the ChatGPT experience that don't always show up in benchmarks but significantly affect whether the tool feels helpful or frustrating in daily use.

What About Thinking Time Settings and Speed Tradeoffs?

OpenAI has been actively tuning thinking time settings for its reasoning models based on user preferences and testing . The company observed that users prefer faster responses, leading to adjustments in Standard and Light thinking time settings. In February 2026, OpenAI made a small reduction to Standard thinking time based on testing, while also restoring the Extended thinking level for GPT-5.2 Thinking to its prior setting after an inadvertent reduction in January.

The thinking level toggle, introduced in September 2025, gives users more choice beyond the default Standard setting. Users can now select Light for faster responses or Extended for more thorough reasoning when depth and accuracy matter more. OpenAI notes that thinking time is not directly comparable across different models, as each is tuned independently to what works best for users. The company plans to continue adjusting these settings as models evolve and will maintain clear user controls when meaningful tradeoffs exist between speed and quality.

What's the Broader Strategy Behind These Updates?

OpenAI's recent releases and updates suggest a shift toward practical, user-focused improvements rather than purely benchmark-driven development . The introduction of GPT-5.3-Codex in February 2026 marked a step-change from code generation to a general-purpose coding agent that users can actively steer while it works. The model is approximately 25 percent faster than its predecessor while setting new highs on key benchmarks.

The rollout of GPT-5.4 mini as a rate limit fallback for paid users demonstrates OpenAI's commitment to maintaining access to reasoning capabilities even during high usage periods. For Plus, Pro, and other paid users, GPT-5.4 mini serves as a fallback when rate limits are reached on GPT-5.4 Thinking, ensuring continued access to advanced reasoning without service interruption. Enterprise customers retain the option to default Auto routing to GPT-5.4 mini if preferred.

These developments indicate that OpenAI is prioritizing the practical deployment of reasoning models in real-world workflows, with emphasis on speed, user control, and reliability. The combination of improved coding, web research, context management, and transparent reasoning processes suggests the company is moving toward AI assistants that can handle complex professional tasks with minimal friction and maximum user agency.