GPT-5.4 Complete Guide: Features & What's New (2026)
OpenAI released GPT-5.4 on March 5, 2026. It is the company's most capable and efficient frontier model for professional work. This guide covers everything you need to know — what changed, who it's for, how to access it, and whether it's worth switching from older models.
What Is GPT-5.4?
GPT-5.4 is a large language model (LLM) built by OpenAI. It unifies the best of their recent work on reasoning, coding, and agentic workflows into one model. It replaces GPT-5.2 Thinking as the default reasoning model in ChatGPT for paid users.
The model comes in three versions:
- GPT-5.4 Thinking — the main reasoning model, available in ChatGPT for Plus, Team, and Pro users
- GPT-5.4 Pro — a higher-performance version for ChatGPT Pro and Enterprise users
- GPT-5.4 (API/Codex) — the developer-facing version with full agentic and computer-use capabilities
It is available in ChatGPT, OpenAI's API, and the Codex programming tool.
What's New in GPT-5.4
Reasoning Gets an Upfront Plan
In ChatGPT, GPT-5.4 Thinking now shows you a plan of its reasoning before it starts answering. You can adjust that plan mid-response. This means fewer follow-up rounds to get the output right.
Native Computer Use
GPT-5.4 is OpenAI's first general-purpose model with native computer-use capabilities. In Codex and the API, it can interact with software applications directly — clicking buttons, filling out forms, and completing tasks across multiple programs. It set an industry record of 75% on the OSWorld-Verified computer use benchmark, edging out human testers who typically score 72.4%.
1 Million Token Context Window
The API version of GPT-5.4 supports up to 1 million tokens of context. Specifically, that's 922,000 input tokens and 128,000 output tokens. This is the largest context window OpenAI has offered. It allows agents to plan and execute long workflows without losing track of earlier information.
Tool Search
GPT-5.4 introduces a new feature called Tool Search. Developers no longer need to manually list every tool their app uses in their API requests. The model can automatically find and call the right tools based on the task. This reduces prompt sizes and cuts inference costs.
Improved Vision
Developers can now upload images with more than 10 million pixels without compressing them. This matters because compression removes detail. The upgrade makes GPT-5.4 significantly better at understanding complex diagrams, screenshots, and high-resolution documents.
Token Efficiency
GPT-5.4 solves the same problems using fewer tokens than GPT-5.2. Fewer tokens means faster responses and lower API costs for developers.
Fewer Hallucinations
OpenAI reports that GPT-5.4 individual responses are 33% less likely to contain errors compared to GPT-5.2. Overall responses are 18% less likely to contain mistakes. That is a meaningful improvement for high-stakes professional use cases.
GPT-5.4 vs GPT-5.2: Key Differences (2026)
| Feature | GPT-5.2 | GPT-5.4 |
|---|---|---|
| Context Window (API) | Not specified | 1 million tokens |
| Computer Use | No | Yes (native) |
| Tool Search | No | Yes |
| Token Efficiency | Baseline | Significantly fewer tokens |
| Error Rate (individual claims) | Baseline | 33% lower |
| Image Input | Compressed only | Up to 10M+ pixels |
| GDPval Score | 70.9% | 83.0% |
| OSWorld-Verified | Below human | 75% (above human) |
Bottom line: GPT-5.4 is the clear upgrade for developers and professional users. GPT-5.2 will remain in ChatGPT as a legacy option through June 5, 2026, then it will be retired.
GPT-5.4 Benchmark Results
These numbers come directly from OpenAI's release and third-party evaluations:
- GDPval (knowledge work across 44 occupations): 83.0% — GPT-5.4 matches or exceeds human professionals in 83% of comparisons, up from 70.9% with GPT-5.2
- OSWorld-Verified (computer use): 75% — above the 72.4% human baseline
- WebArena Verified (web navigation): record score
- Investment banking benchmark (financial modeling): 87.3% with GPT-5.4 Thinking, up from 43.7% with the original GPT-5
These aren't just synthetic scores. The GDPval benchmark tests real tasks: generating sales presentations, building accounting spreadsheets, creating manufacturing diagrams, and drafting urgent care schedules.
Who Should Use GPT-5.4
GPT-5.4 is built for professional and enterprise workflows. It's particularly useful for:
Developers and engineers who need agents that can run across long code sessions, use external tools without manual setup, and interact with desktop software.
Finance professionals who work with financial models, scenario analysis, and data-heavy research. OpenAI's internal investment banking benchmark showed performance nearly doubled compared to the original GPT-5.
Knowledge workers across industries like law, consulting, healthcare, and marketing who need to produce documents, presentations, and reports efficiently.
Researchers who work with large documents or multi-source analysis that exceeds what older context windows could hold.
How to Access GPT-5.4
In ChatGPT:
- Log into ChatGPT at chat.openai.com
- Make sure you have a Plus, Team, Pro, or Enterprise plan
- Select GPT-5.4 Thinking from the model picker
- GPT-5.4 Pro is only available on Pro and Enterprise plans
In the API:
The model is available as gpt-5.4 in both the Responses API and Chat Completions API. The 1 million token context window is only available via the API, not in ChatGPT's standard interface.
In Codex:
GPT-5.4 is rolling out across all Codex surfaces for paid users. It replaces GPT-5.2 as the default reasoning model in Codex.
Note: GPT-5.2 Thinking remains available under "Legacy Models" in the ChatGPT model picker through June 5, 2026. Enterprise and Edu users can enable GPT-5.4 early via admin settings.
How GPT-5.4 Works: The Key Techniques
Chain-of-Thought with Steering
GPT-5.4 uses chain-of-thought reasoning — it works through problems step by step rather than jumping to a final answer. What is new is that ChatGPT now exposes this process to you upfront as a plan. You can redirect it before it finishes.
Agentic Tool Use
The model is designed to act as an agent. That means it can call external tools, browse the web, interact with software, and chain multiple actions together to complete a task — all without you manually managing each step.
Codex Integration
GPT-5.4 is the first general model that brings GPT-5.3-Codex's coding capabilities into the main model line. Previously, coding power and general reasoning lived in separate models. Now they're combined.
Tips for Getting the Best Results
- Use the thinking plan: When you see the upfront plan in GPT-5.4 Thinking, read it before the model finishes. Redirect it early rather than correcting the final output.
- Leverage the 1M context via API: If you're working with long documents or complex codebases, use the API rather than ChatGPT to access the full context window.
- Skip tool lists with Tool Search: If you're building on the API, let Tool Search handle tool discovery instead of listing every tool manually. This cuts prompt size and speeds up responses.
- Upload full-resolution images: Don't compress images before sending them. GPT-5.4 can now handle 10 million+ pixel images, so you won't lose detail.
- Chain tasks for complex workflows: GPT-5.4 is built for multi-step tasks. Give it a full goal — not just a single-step prompt — and let it plan the workflow.
Common Mistakes to Avoid
Assuming GPT-5.4 is available for free users. It isn't. Free ChatGPT users do not have access to GPT-5.4 Thinking. Free users fall back to GPT-5-mini after hitting usage limits.
Using GPT-5.4 for simple tasks. For basic questions and short responses, a faster and cheaper model may serve you better. GPT-5.4 is designed for complex, multi-step professional work — don't pay for a sledgehammer when you need a nail.
Not using the context window for long projects. Developers who stay at shorter context lengths miss out on one of GPT-5.4's biggest advantages. Plan your architecture to take advantage of the 1 million token window where it matters.
Ignoring Tool Search. Manually maintaining long tool lists in your API prompts adds cost and overhead. Tool Search was built to replace that pattern.
Frequently Asked Questions
What is GPT-5.4 and when was it released?
GPT-5.4 is OpenAI's latest frontier model, released on March 5, 2026. It unifies reasoning, coding, and agentic capabilities into one system. It replaces GPT-5.2 Thinking as the default in ChatGPT for paid users and is also available in the OpenAI API and Codex.
What plans include GPT-5.4 in ChatGPT?
GPT-5.4 Thinking is available for ChatGPT Plus, Team, and Pro users. GPT-5.4 Pro requires a Pro or Enterprise plan. GPT-5.4 is not available to free ChatGPT users. Enterprise and Edu plans can enable early access through admin settings.
How is GPT-5.4 different from GPT-5.2?
GPT-5.4 adds native computer-use capabilities, a 1 million token context window in the API, Tool Search for automatic tool discovery, and improved image input. It is also 33% less likely to make errors on individual claims than GPT-5.2.
Can GPT-5.4 control my computer?
In the API and Codex, yes. GPT-5.4 has native computer-use capabilities, meaning it can interact with software interfaces, click elements, and carry out workflows across applications. This feature is not available in the standard ChatGPT interface.
Is GPT-5.4 available via the OpenAI API?
Yes. GPT-5.4 is available in both the Responses API and Chat Completions API. The API version supports a 1 million token context window and includes Tool Search and full computer-use capabilities. GPT-5.4 Thinking is also available as a Pro tier via the API.
Will GPT-5.2 still work after GPT-5.4 launches?
Yes, but not for long. GPT-5.2 Thinking will remain available in ChatGPT under Legacy Models until June 5, 2026, when it will be retired. GPT-5.1 and GPT-5 remain available in the API with no current deprecation plans.
Conclusion
GPT-5.4 is a meaningful leap forward from GPT-5.2. It brings native computer use, a 1 million token context window, smarter tool use, and significantly fewer errors — all in one model. The addition of an upfront reasoning plan in ChatGPT gives users more control over complex tasks without extra back-and-forth.
For developers, the Tool Search feature and expanded vision capabilities lower the cost and complexity of building agentic applications. For professionals in finance, law, or consulting, the benchmark results on real-world tasks are compelling.
If you're on a paid ChatGPT plan, GPT-5.4 Thinking is your new default. If you're building on the API, it's worth migrating now. GPT-5.2 has a sunset date — GPT-5.4 is where the roadmap points.