OpenAI has officially released GPT-5.4, a massive update that brings new levels of intelligence to everyday tasks. With the launch of the ChatGPT 5.4 thinking model, OpenAI introduces advanced computer use and vision, seamless Excel integration, and deep data plugins. Whether you need to automate financial models, write complex code, or simplify research and analysis, GPT-5.4 is designed to handle serious professional and academic work.
Nonetheless, this power comes with a premium price tag, setting up a major battle with rival AI models.
Key Takeaways:
- Built for Finance and Spreadsheets: OpenAI launched a native “ChatGPT for Excel” beta. It connects to live data from FactSet and S&P Global, letting users build financial models using plain English.
- Record-Breaking Performance: The new GPT-5.4 Thinking model dominates the ChatGPT 5.4 benchmark tests. It scores higher than older models in law, coding, and general knowledge work.
- Agentic Computer Use: GPT-5.4 does not just chat. It can take control of software, navigate websites, and execute multi-step tool workflows faster and cheaper than before.
What does GPT-5.4 do? The Power of ‘Thinking’
Advanced Reasoning and Logic
If you are wondering, what does GPT-5.4 do?
The answer lies in its ability to think before it acts. The GPT-5.4 Thinking model uses advanced abstract reasoning to solve ambiguous problems. It introduces human-like computer interaction, allowing it to execute tasks inside your existing systems without needing external tools.
Improved Web Search and Context
It also features improved web search and an incredibly long context window. This means it can read massive documents, like legal contracts or entire codebases, without forgetting details.
“GPT-5.4 sets a new bar for document-heavy legal work. On our BigLaw Bench eval, it scored 91%. Compared to other models, GPT-5.4 is currently better at structuring complex transactional analysis, maintaining accuracy across lengthy contracts, and delivering the high level of detail legal practitioners require.”
— Niko Grupen, Head of Applied Research at Harvey. [Source: OpenAI]
GPT-5.4 loves Spreadsheets: The Finance Revolution
ChatGPT for Excel Integration
It is no secret that GPT-5.4 loves spreadsheets.
OpenAI has officially entered the financial software market by releasing ChatGPT for Excel in beta. Users no longer need to manually type every formula. They can talk to their spreadsheets in plain English. This allows them to create or change live models automatically.
Real-Time Data Plugins
“On internal finance and Excel evaluations, GPT-5.4 improved accuracy by 30 percentage points, which he links to expanded automation for model updates and scenario analysis.”
— Daniel Swiecki, Walleye Capital [Source: VentureBeat]
Users can now pull live market data from trusted providers like FactSet, S&P Global, and LSEG. This helps investment teams quickly conduct due diligence and value companies. Instead of typing formulas manually, you just describe the scenario in plain English, and the AI builds the formatted model.
“ChatGPT has materially accelerated our research and due diligence workflows from financial analysis and market research to legal review and writing internal memos while improving consistency across teams. It has expanded our team’s capacity, freeing our investment professionals to focus more time on judgment, debate, and conviction. We’re excited to be early adopters of new capabilities and to help shape how AI transforms financial services in the years ahead.”
— Amr Ellabban, PhD, Head of AI, Hg. [Source: OpenAI]
ChatGPT 5.4 Benchmark and Evaluations
Breaking Records in Knowledge Work
OpenAI provided extensive evaluations showing that GPT-5.4 beATS its predecessor, GPT-5.2. On the GDPval test, which measures professional knowledge work, GPT-5.4 matched or beat human industry professionals 83% of the time (up from 70.9%).
Dominating Professional Services
The model also excels in specialized fields. In internal tests simulating the work of junior investment bankers, GPT-5.4 scored an impressive 87.3% on spreadsheet modeling.
“GPT-5.4 is the best model we’ve ever tried. It’s now top of the leaderboard on our APEX-Agents benchmark, which measures model performance for professional services work. It excels at creating long-horizon deliverables such as slide decks, financial models, and legal analysis, delivering top performance while running faster and at a lower cost than competitive frontier models.”
— Brendan Foody, CEO at Mercor. [Source: OpenAI]
Computer use and vision: A New Agentic Era
Unmatched Multi-Step Tool Use
GPT-5.4 takes tool use to the next level. It can navigate software interfaces on its own.
“GPT-5.4 xhigh is the new state of the art for multi-step tool use. Zapier runs some of the most rigorous tool use benchmarks in the industry, testing models across hundreds of advanced real-world workflows. GPT-5.4 finished the job where previous models gave up – the most persistent model to date.”
— Wade, CEO at Zapier. [Source: OpenAI]
Navigating the Web Autonomously
Its computer vision allows it to “see” and interact with websites like a human.
“In our evals measuring computer use performance across ~30K HOA and property tax portals, GPT-5.4 achieved a 95% success rate on the first attempt and 100% within three attempts… It also completed sessions ~3x faster while using ~70% fewer tokens.”
—- Dod Fraser, CEO at Mainstay. [Source: OpenAI]
Coding: GPT 5.4 vs GPT 5.3 Codex
The Ultimate Developer Assistant
For programmers, Coding has seen a massive upgrade. Developers are using GPT 5.4 cursor integrations to write and audit code rapidly. This tool proactively parallelizes work. It keeps things moving swiftly.
Assertiveness and Speed
When looking at GPT 5.4 vs GPT 5.3 codex, the new GPT 5.4 benchmarks show a distinct personality shift. It is much smarter and more confident.
“GPT-5.4 is currently the leader on our internal benchmarks. Our engineers find it to be more natural and assertive than previous models. It works through ambiguous problems without second-guessing itself, and it’s proactive about parallelizing work to keep things moving.”
— Lee Robinson, VP of Developer Education at Cursor. [Source: OpenAI]
Availability and pricing: Is GPT-5.4 free?
Targeting Premium Users
Many users are asking, Is GPT-5.4 free?
The answer is no.
The heavy focus on coding and professional tasks is clear. This focus shows that OpenAI is targeting paying professionals. They are directly competing with Anthropic’s expensive Claude Opus 4.6.
Two Tiers of Power and API Costs
To access it, you need a paid plan:
- GPT-5.4 Thinking: Available to all ChatGPT Plus ($20/month) subscribers.
- GPT-5.4 Pro: Reserved for the new ChatGPT Pro ($200/month) and Enterprise users.
- API Pricing: GPT-5.4 costs $2.50 per 1M input tokens and $15 per 1M output tokens. The Pro version costs $30 for input and $180 for output.
GPT-5.4 Thinking System Card and Safety
Documenting Safety Limits
With great power comes the need for rigorous safety protocols. OpenAI released the GPT-5.4 Thinking System Card to publicly document how the model handles complex tasks securely.
Managing Concurrent Processes
OpenAI optimizes workflow and tracks how the model “thinks.” This ensures it behaves reliably and safely. It manages concurrent processes, accesses live financial data, and executes independent computer use tasks on behalf of the user.
Learn More: GPT‑5.4 Thinking System Card
Community Reactions on OpenAI GPT 5.4 model: The Verdict from Real Users
Positive Reactions
Users are amazed by the speed, vision, and coding accuracy of the new model.
“AI vision is catching up to humans faster than expected. On the EyeBench-V2 benchmark: Human baseline: 100% GPT-5.4 Pro: 90% GPT-5.4: 71% … Most other models are far behind.”
“I am still evaluating GPT 5.4 but it has the speed of 5.3-codex (5.4 feels faster).”
“Using 5.4-high to do the work, and 5.3-codex to review/audit. Followed the plan completely, and reviews came up squeaky clean. OpenAI’s pace of development is insane.”
Negative Reactions
However, some users report bugs, lazy assumptions, and frustration with the model’s logic.
“For me it’s shit, it got everything I asked for wrong except documentation – It kept getting everything wrong that I changed back to 5.3-codex… maybe it’s just me”
“I asked it to fix a pretty easy bug and it took the easiest path making assumptions that were not correct and did not bother to check anything else. It told me that a sql table schema must have changed and added logic to drop the table and recreate it which would have been devastating if I implemented the code. It seems lazy and a little dangerous. Back to 5.2 for me.”
“GPT 5.4 delivers unreliable reasoning, lack of common sense, and ideological biases that render interactions with this model boring and flakey. OpenAI successfully lobotomised its latest release.”
Action Points — How to start using GPT-5.4?
- Upgrade Your Plan: To access the new features, make sure you are subscribed to ChatGPT Plus ($20/month) or Pro ($200/month).
- Install the Excel Add-in: If you work in finance, download the ChatGPT for Excel beta add-in. Start pulling live data from S&P Global and FactSet directly into your spreadsheets.
- Test Computer Use: Try giving the AI a multi-step task. Ask it to research data across several websites. Then, paste the results into a document to test its new agentic tool use.
FAQs on GPT-5.3-Codex
- Is GPT-5.4 free?
No, it is currently reserved for paid subscribers (Plus, Pro, and Enterprise).
- How to use GPT-5 4?
You can select it from the model dropdown in your ChatGPT Plus or Pro account. You can also use it via the API. Alternatively, access it through the new ChatGPT for Excel add-in.
- What does GPT-5.4 do?
It performs advanced abstract reasoning, handles complex coding, uses computer tools autonomously, and integrates deeply with Excel for financial modeling.
- What is the GPT-5.4 Thinking System Card?
It is a document published by OpenAI outlining the safety, testing, and operational limits of the new model.
- How does the ChatGPT 5.4 benchmark compare to older models?
It significantly outperforms GPT-5.2 in professional tasks, scoring 83% on GDPval and 91% on BigLaw Bench.
- Does GPT-5.4 love spreadsheets?
Yes! The model has native integration with Microsoft Excel and connects to live market data providers for fast financial modeling.
- What is the difference between GPT-5.4 vs 5.3 codex?
Users report that GPT-5.4 is faster, handles parallel tasks better, and is more assertive when writing code compared to the 5.3 Codex model.
- Does it have computer vision?
Yes, its computer use and vision capabilities are top-tier, scoring 90% on the EyeBench-V2 test for the Pro model.
- Who is GPT-5.4 designed for?
It is designed for professionals doing deep knowledge work, coding, legal analysis, and academic research.
- How much does the API cost?
The standard 5.4 API costs $2.50 per 1M input tokens and $15 per 1M output tokens.
- Can GPT-5.4 handle multi-step workflows automatically?
Yes, it is highly capable of completing long-horizon tasks. Using the GPT-5.4 xhigh setting, Zapier rigorously tested it. They praised it as the most persistent model to date. It finishes complex workflows where previous AI models gave up.
- How does the model perform on legal documents?
It is setting a new standard for the legal industry. It scored 91% on the BigLaw Bench evaluation. It is specifically tuned to maintain accuracy and detail across lengthy contracts. It also excels in complex transactional analysis.
- Which financial data providers connect to ChatGPT for Excel?
The new beta add-in introduces deep data integrations with major financial providers. These include FactSet, S&P Global, and LSEG (London Stock Exchange Group). This allows you to pull real-time data directly into your workbook.
- How reliable is its “computer use” in real-world scenarios?
It is incredibly reliable. In tests navigating over 30,000 different Homeowner Association (HOA) and property tax portals, GPT-5.4 achieved a 100% success rate within three attempts, operating 3x faster than older computer-use models.
- Does the model require external tools for computer interaction?
No, one of its biggest upgrades is advanced human-like computer interaction (CUA). This allows GPT-5.4 to execute tasks directly within your existing systems and software interfaces without needing external, third-party automation tools.
Further Reading
- Official OpenAI Announcement: Introducing GPT-5.4
- ChatGPT for Excel: Learn more about the integration
- Safety and Security: GPT-5.4 Thinking System Card
- Latest Model Resources: OpenAI Academy
Check out more ChatGPT Hacks on AppliedAI Tools
Make the most of ChatGPT for your personal or workplace productivity:
- How to Use ChatGPT After Limit Reach?
- How to Use ChatGPT Without SignUp [Privacy Tips]
- ChatGPT for Daily Morning Routine Cheat Sheet
- How to Use ChatGPT with Canva for Instant Designs
- 4 ChatGPT Prompts To Find Discount Coupon Codes
- Learn English Using ChatGPT — Free AI Language Tutor
- How to Make ChatGPT Forget Memory?
- ChatGPT For Free Personal Color Analysis – Prompts and GPTs
- How to use ChatGPT on WhatsApp in 3 steps
- How To Digitize Handwritten Notes with ChatGPT For Free
- Claim ChatGPT Go Free in India: Features and Reviews
- 5 ChatGPT Parental Controls: OpenAI Age Detection Tech Explained
- How OpenAI Critterz Makes Filmmaking Accessible To Independent Animators
- Use ChatGPT Projects Feature For Free [Tutorial with Example]
- Best AI Browsers: Free Comparison Business Analysis Report
- ChatGPT Search Upgrade: Shopping Without Login
Twice a month, we share AppliedAI Trends newsletter.
Get SHORT AND ACTIONABLE REPORTS on AI Trends across new AI tools launched and jobs affected due to AI tools. Explore new business opportunities due to AI technology breakthroughs. This includes links to top articles you should not miss, like this ChatGPT hack tutorial you just read.
Subscribe to get AppliedAI Trends newsletter – twice a month, no fluff, only actionable insights on AI trends:
You can access past AppliedAI Trends newsletter here:
This blog post is written using resources of Merrative. We are a publishing talent marketplace that helps you create publications and content libraries.
Get in touch if you would like to create a content library like ours. We specialize in the niche of Applied AI, Technology, Machine Learning, or Data Science.
Leave a Reply