To use ChatGPT 5.4, first choose where you want to work. The API still lists GPT-5.4, but ChatGPT’s current paid experience uses GPT-5.6 Sol, with GPT-6 Pro on eligible plans. GPT-6 Pro is powered by GPT-6 Astra. Start with the current model picker and plan options if you want the latest ChatGPT experience.
Subscribing directly often means high costs, strict usage limits, and region restrictions. GlobalGPT brings 100+ models into one account for text, image, and video work. BASIC starts at $5.8 per month billed annually, and PRO is $10.8 per month billed annually at the prices checked on September 13, 2026. GPT-6 Astra is available in its model menu.
Choose GPT-6 Astra on GlobalGPT for demanding writing and reasoning in that shared workspace. Model use draws on the selected plan’s credits; native ChatGPT, Work, and Codex features have their own product access rules.
What is ChatGPT 5.4? Understanding OpenAI’s 2026 Frontier Model
Released on March 5, 2026, GPT-5.4 brought reasoning, coding, and agentic workflows together for professional work. It is now an earlier model in OpenAI’s model catalog; GPT-6 Astra is the current flagship.
GPT-5.4 was designed for tasks that combine research, coding, and work across several tools.
About the comparisons below: The GPT-5.4 benchmark results and feature examples come from its March 2026 release. They describe that version and its test setup; they are not scores for GPT-6 Astra.
The Evolution: From GPT-5.2 to GPT-5.4’s Agentic Workflows
| Benchmark | GPT-5.4 | GPT-5.3-Codex | GPT-5.2 |
|---|---|---|---|
| GDPval (wins or ties) | 83.0% | 70.9% | 70.9% |
| SWE-Bench Pro (Public) | 57.7% | 56.8% | 55.6% |
| OSWorld-Verified | 75.0% | 74.0%* | 47.3% |
| Toolathlon | 54.6% | 51.9% | 46.3% |
| BrowseComp | 82.7% | 77.3% | 65.8% |
*Previously reported as 64.7%. GPT‑5.3‑Codex achieves 74.0% with a newly introduced API parameter that preserves the original image resolution.
Building on GPT-5.2’s reasoning capabilities, GPT-5.4 adds capabilities for agentic workflows. It incorporates the coding capabilities of GPT-5.3-Codex while improving how the AI interacts with software environments and tools.
Notably, it reduces hallucinations: individual claims are 33% less likely to be false, and full responses are 18% less likely to contain errors relative to GPT-5.2.
ChatGPT 5.4 Thinking vs. GPT-5.4 Pro vs. Codex
GPT-5.4 appeared in three different product contexts at launch. For access today, use the current choices in the table below.
- GPT-5.4 Thinking: Replaced GPT-5.2 Thinking at launch in ChatGPT, providing upfront plans and advanced web research capabilities.
- GPT-5.4 Pro: Designed for users who need maximum performance on the most complex, multi-step tasks.
- GPT-5.4 (Codex/API): The first general-purpose model with native computer-use capabilities, supporting up to a 1M token context window for long-horizon planning.
| Where you work | Current choice | What to use it for | Access |
|---|---|---|---|
| ChatGPT | GPT-5.6 Sol | Instant, Medium, or High; Extra High depends on plan | Plus includes Medium/High; Pro, Business, and Enterprise add Extra High and Pro options |
| ChatGPT Pro models | GPT-6 Pro; GPT-5.6 Sol Pro | Harder questions and longer reasoning | Eligible Pro, Business, and Enterprise plans; workspace settings may apply |
| Work / Codex | GPT-6 Astra; GPT-5.6 Sol, Terra, Luna | Files, coding, and tasks with tools | Astra is included in Plus and Pro in Work/Codex; allowances are separate from Chat |
| OpenAI API | gpt-5.4 / gpt-5.4-pro | A specific GPT-5.4 integration | Separate API billing; Pro uses the Responses API |
How to Access and Enable ChatGPT 5.4 in Your Account
For the current ChatGPT experience, choose the model available to your plan. If you specifically need GPT-5.4, use its documented API model ID or an available legacy option in your account; do not assume every paid account still shows the March 2026 menu.
Selecting Models in ChatGPT Plus, Business, and Pro Plans
ChatGPT Plus costs $20 per month, billed monthly. It includes the GPT-5.6 Sol Medium and High reasoning options, but not the Pro model option in standard Chat. If you are comparing regional checkout totals, see how much ChatGPT Plus costs.
Open the model picker, choose GPT-5.6 Sol, and select the reasoning level you need. GPT-6 Pro is available on eligible Pro, Business, and Enterprise plans. Plus users can use GPT-6 Astra in Work and Codex, which have separate usage allowances.
- Sign in to the account that holds your subscription.
- Open the model picker. Use Instant for short questions or Medium/High for multi-step work.
- Add your task, source material, output format, and any constraints.
- If a model is missing, check your plan, workspace permissions, and the app version before buying another subscription.
API Setup for Developers and Context Window Configurations
Developers can set model: "gpt-5.4" or model: "gpt-5.4-pro" in a request to POST /v1/responses. These are model IDs, not endpoint URLs. GPT-5.4 Pro uses the Responses API only.
GPT-5.4 has a 1,050,000-token API context window and a 128,000-token maximum output. In Codex, model_context_window and model_auto_compact_token_limit are configuration settings, not fields to paste into a Responses API request. See the GPT-5.4 API specification for its limits.
from openai import OpenAI
client = OpenAI()
response = client.responses.create(
model="gpt-5.4",
input="Turn these meeting notes into an action list with owners and deadlines.",
reasoning={"effort": "medium"}
)
print(response.output_text)
Step-by-Step Guide: How to Use the ChatGPT 5.4 “Thinking” Features
For longer tasks, ask for a short plan before the final answer. In current ChatGPT, eligible paid users can choose GPT-5.6 Sol Medium or High; Pro, Business, and Enterprise also include Extra High. The available reasoning controls depend on your plan.
Reading the Upfront Plan and Preamble
At launch, GPT-5.4 Thinking introduced a short preamble for longer, more complex queries. It outlined the intended approach before the final answer.
Read the plan before committing to a long answer. If the scope is wrong, add the missing audience, budget, deadline, or source restriction.
Try this prompt: “Use only the notes below. First list the decisions still needed. Then draft a one-page launch plan with an owner, deadline, and cost for each task. Flag missing information instead of filling it in.”
Mid-Response Course Correction: Steering Output in Real-Time
GPT-5.4 Thinking introduced the option to add instructions while a response was still in progress. In a current session, use that control if it is available in your interface.
Add a specific correction, such as “focus on the European market” or “keep the budget below $5,000,” then check that the final answer follows it.
If you are new to the interface, the ChatGPT getting-started guide covers chats, uploads, and follow-up prompts before you tackle longer workflows.
Mastering Native Computer Use and Vision in GPT-5.4
GPT-5.4 marks a major milestone as OpenAI’s first general-purpose model with native, state-of-the-art computer-use capabilities.
Operating Browsers and Software via Codex and the API
Developers can build agents that complete real tasks across software systems by issuing mouse and keyboard commands in response to screenshots.
Computer use needs a tool-enabled environment that can return screen images and execute actions. A normal chat response alone does not operate your desktop. Keep consequential actions behind the confirmation controls of the product you use.
In OpenAI’s March 2026 OSWorld-Verified results, GPT-5.4 achieved a 75.0% success rate, compared with GPT-5.2’s 47.3% and the benchmark’s human baseline of 72.4%.
Using “Playwright Interactive” for Visual Debugging and App Testing
OpenAI released an experimental Codex skill called Playwright (Interactive). The skill lets Codex inspect web and Electron applications while debugging them.
OpenAI demonstrated this with an isometric theme park game that Codex could build and playtest during development.
Reading Small Text and Detailed Documents
For dense images, the original input detail level supports up to 10.24 million total pixels or a 6,000-pixel maximum dimension, whichever limit is reached first. That extra detail can help with small text and precise visual targets.
On OmniDocBench, GPT-5.4 recorded an average error rate of 0.109, a noticeable improvement over GPT-5.2.
MMMUPro was run with reasoning effort set to xhigh. OmniDocBench was run with reasoning effort set to none, to reflect low-cost, low-latency performance.
Boosting Professional Productivity with ChatGPT 5.4 Tools
GPT-5.4 was built to execute real work products, from urgent care schedules to manufacturing diagrams. It matched or beat professional reference work in 83.0% of GDPval comparisons. For business adoption, also review the terms for ChatGPT commercial use in 2026.
What the GDPval Results Measure
Building on GPT-5.2’s general reasoning capabilities, GPT-5.4 delivers more consistent and polished results on real-world tasks critical to professionals. Useful examples include schedules, diagrams, and documents with several requirements.
On the GDPval benchmark, which tests an agent’s ability to produce well-specified knowledge work across 44 distinct occupations, GPT-5.4 improved on GPT-5.2 in the launch evaluation. It matched or exceeded industry professionals in 83.0% of comparisons, a substantial jump from GPT-5.2’s 70.9% rate. This demonstrates GPT-5.4’s superior capacity for high-level professional automation.
Installing and Using the New ChatGPT for Excel Add-in
For financial professionals and data analysts, OpenAI has launched the official ChatGPT for Excel add-in.
For a quick spreadsheet task, uploading a workbook is a separate route from installing an add-in. The guide to using Excel files in ChatGPT walks through the upload workflow. Start with a copy of the workbook and ask the model to explain every changed formula.
When tested on internal spreadsheet modeling tasks typical for a junior investment banking analyst, GPT-5.4 achieved a mean score of 87.3%, compared with the previous 68.4% score.
The 87.3% figure is from OpenAI’s detailed evaluation table for GPT-5.4. It is a benchmark score, not a guarantee that a financial model will be correct without review.
Automating Presentations, Legal Contracts, and Complex Documents
GPT-5.4 was also evaluated on longer document and presentation tasks. In presentation tasks, human raters preferred GPT-5.4’s output 68.0% of the time due to stronger aesthetics and effective use of image generation.
In OpenAI’s legal evaluation, GPT-5.4 scored 91% on BigLaw Bench, which tested complex legal work.
API Developer Guide: Tool Search and the 1M Context Window in GPT-5.4
For developers managing large ecosystems of tools, GPT-5.4 can load tool definitions as needed to reduce context use.
How the New “Tool Search” Saves Tokens and Lowers Latency
Previously, all tool definitions had to be included upfront, wasting thousands of tokens. The new Tool Search feature allows the model to receive a lightweight list and look up specific tool definitions only when needed.
Loading fewer definitions can also reduce latency. On Scale’s MCP Atlas benchmark, the tool-search configuration successfully reduced total token usage by 47% without sacrificing any accuracy.
Configuring the 1 Million Token Window for Long-Horizon Tasks
When deploying agents for long-horizon deliverables, managing context is critical. The 1M token context window allows agents to keep more source material available during a task.
The API can accept large code or document collections, subject to the model’s input and output limits.
Start with the smallest useful document set. A larger window can hold more material, but it does not make every task cheaper. If you are deciding whether to move to a newer model, compare GPT-5.6 pricing and billing before expanding the workload.
ChatGPT 5.4 Pricing, Cost Efficiency, and Token Limits
Because GPT-5.4 is highly token-efficient, it uses significantly fewer tokens to solve problems compared to GPT-5.2. Understanding the full breakdown of GPT-5.4 pricing is crucial for developers optimizing these long-horizon workflows.
Token Usage, Batch Pricing, and Speed Improvements vs. Legacy Models
While the base price per token is higher to reflect its advanced capabilities, the total cost depends on both the token rate and the number of tokens used.
| API model | Input / 1M tokens | Cached input / 1M | Output / 1M |
|---|---|---|---|
| gpt-5.4 | $2.50 | $0.25 | $15.00 |
| gpt-5.4-pro | $30.00 | — | $180.00 |
| gpt-6-astra | $10.00 | $1.00 | $50.00 |
| gpt-5.6-sol | $4.00 | $0.40 | $20.00 |
| gpt-5.6-terra | $2.00 | $0.20 | $12.00 |
| gpt-5.6-luna | $0.20 | $0.02 | $1.20 |
Source: OpenAI API pricing. These are API charges, billed separately from ChatGPT subscriptions. For GPT-5.4 and GPT-5.4 Pro, prompts above 272K input tokens cost 2× the input rate and 1.5× the output rate for the full session. GPT-5.4 Batch and Flex discounts are separate from the standard rates shown here.
For GPT-5.4, that long-context rate is $5 input and $22.50 output per million tokens; GPT-5.4 Pro is $60 input and $270 output. Current-model long-context rates, cache writes, tools, and regional processing have their own lines on the official pricing page.
ChatGPT subscriptions and GlobalGPT plans
| Plan | Price and billing | What matters before paying |
|---|---|---|
| ChatGPT Plus | $20/month, billed monthly | GPT-5.6 Sol Medium/High; API usage is separate |
| ChatGPT Pro $100 | $100/month | Available for new subscriptions; Pro-model allowances apply |
| ChatGPT Pro $200 | $200/month for existing subscriptions | New subscriptions and upgrades paused from September 10, 2026 |
| GlobalGPT BASIC | $5.8/month, billed annually | 144,000 credits/year on the displayed plan |
| GlobalGPT PRO | $10.8/month, billed annually | 240,000 credits/year on the displayed plan |
ChatGPT Plus is $20 per month and does not include API usage. That is a monthly subscription price, not a price for one million tokens.
OpenAI has paused new Pro $200 subscriptions and upgrades since September 10, 2026. Existing Pro $200 subscriptions continue to renew, and Pro $100 remains available.
The GlobalGPT annual plans combine multiple model families in one workspace. Credits are generally allocated and reset monthly; unlimited usage applies only to models specifically marked for unlimited access.
Safety, Privacy, and ZDR: Is ChatGPT 5.4 Secure for Enterprise?
OpenAI maintains a precautionary approach to deployment, treating GPT-5.4 as a “High cyber capability” model under their Preparedness Framework.
Chain-of-Thought (CoT) Controllability and Cyber Risk Mitigation
OpenAI also studied GPT-5.4’s low Chain-of-Thought (CoT) controllability. In the reported tests, the model had limited ability to deliberately alter its chain of thought. That finding does not make monitoring a complete safety guarantee.
For eligible API customers, Zero Data Retention requires OpenAI approval and has endpoint-specific limits. Standard API abuse-monitoring logs can be retained for up to 30 days, with exceptions described in the policy. Do not treat a model choice or a ChatGPT subscription as a blanket ZDR guarantee.
If you use more than one provider, compare AI data privacy policies before uploading client work. The service you send a file to determines the applicable retention and account controls.
ChatGPT 5.4 FAQ
Can I still use ChatGPT 5.4?
GPT-5.4 is still documented in the OpenAI API. In ChatGPT, use the models available in your account; the current paid experience uses GPT-5.6 Sol, with GPT-6 Pro on eligible plans.
Is GPT-5.4 the latest OpenAI model?
No. As of September 13, 2026, GPT-6 Astra is OpenAI’s current flagship. GPT-5.6 Sol, Terra, and Luna are also current options. GPT-5.4 remains a distinct earlier model.
Which ChatGPT plan do I need for advanced reasoning?
Plus includes GPT-5.6 Sol Medium and High. Pro, Business, and Enterprise add Extra High and Pro options. GPT-6 Pro access and limits depend on the plan and, for managed accounts, workspace settings.
How much does GPT-5.4 cost through the API?
Standard short-context rates are $2.50 input, $0.25 cached input, and $15 output per million tokens. GPT-5.4 Pro is $30 input and $180 output. Long-context and tool charges are separate.
Does ChatGPT Plus include API credits?
No. ChatGPT Plus costs $20 per month and API usage is billed separately. A subscription to ChatGPT, an OpenAI API account, and a GlobalGPT subscription have different billing arrangements.
Can I buy the ChatGPT Pro $200 plan now?
New Pro $200 subscriptions and upgrades have been paused since September 10, 2026. Existing subscriptions continue to renew. Pro $100 remains available; the pause has no announced end date in the checked notice.
Can I use current OpenAI models on GlobalGPT?
Yes. GlobalGPT has verified entries for GPT-6 Astra and GPT-5.6 Sol. Its BASIC and PRO annual plans display $5.8 and $10.8 per month. Model use follows GlobalGPT’s credit rules and does not include every native OpenAI product feature.
Does the 1M context window apply to ordinary ChatGPT chats?
Do not assume it does. GPT-5.4’s API specification lists 1,050,000 context tokens and 128,000 maximum output tokens. ChatGPT and Codex have separate context, access, and usage rules.
Conclusion: Maximizing Your Workflow and Agents with ChatGPT 5.4
Use GPT-5.4 when you need that specific API model or want to understand its reasoning and tool workflows. For the latest ChatGPT experience, use the current plan options; for a shared multi-model workspace, start with GPT-6 Astra on GlobalGPT. Begin with one useful task, check the result, and increase the workload only when the output earns it.
