OpenAI Pro model
GPT-5.4 Pro
A higher-compute GPT-5.4 variant for long-running, difficult problems that benefit from deeper reasoning, large context, and tool-driven execution.
- Context window
- 1.05M
- tokens
- Max output
- 128K
- tokens
- Input
- $30.00
- per 1M tokens
- Cached input
- No discount
- same token rate
- Output
- $180.00
- per 1M tokens
01 / Overview
What GPT-5.4 Pro Is
GPT-5.4 Pro is OpenAI's higher-compute version of GPT-5.4, designed for tough problems that benefit from longer reasoning and more consistent answers rather than the lowest possible latency or cost.
A Specialized Tier for Problems That Need More Thought
Standard GPT-5.4 is positioned as a general-purpose frontier model for coding and professional work. GPT-5.4 Pro takes a different role: it allocates more compute to difficult requests in order to improve precision and consistency.
That distinction matters in production. A Pro model is not automatically the right default for every task. Its value appears when the application contains a meaningful class of problems that cheaper or faster models solve inconsistently, require multiple retries, or send back for human review.
GPT-5.4 Pro is especially notable because it combines deep reasoning with a large context window and a targeted agentic toolset. It can search the web and files, use a computer interface, apply patches, work through MCP integrations, and discover tools through Tool Search.
OpenAI notes that some requests can take several minutes to complete and recommends background mode for long-running calls. This makes GPT-5.4 Pro better understood as a high-value problem solver than as a low-latency conversational model.
- Higher-compute version of GPT-5.4.
- Designed for tougher problems and more consistent answers.
- Supports a 1.05M-token context window.
- Uses medium, high, or xhigh reasoning effort.
- Best evaluated on hard cases where improved precision has measurable value.
- Provider
- OpenAI
- Family
- GPT-5.4
- Tier
- Pro
- Knowledge cutoff
- Aug 31, 2025
- Input modalities
- Text, Image
- Output modality
- Text
- Default reasoning
- Medium
02 / Use cases
Where GPT-5.4 Pro Fits Best
GPT-5.4 Pro is most appropriate for long-running tasks where the quality of the reasoning path matters more than immediate response time and where a stronger answer can offset premium inference cost.
Use Pro as an Escalation Layer
A useful deployment pattern is to reserve GPT-5.4 Pro for requests that cross a complexity threshold. Normal traffic can run on a general-purpose or lower-cost model, while difficult cases are escalated when the task contains more uncertainty, higher business value, or a greater cost of failure.
Technical investigations are one example. The model may need to inspect a large body of evidence, reason across code and documentation, identify the likely failure mode, and then use tools to move toward a fix.
Long-form professional analysis is another. A task can require reconciling several documents, preserving constraints across a large context, and producing a conclusion that remains internally consistent after substantial reasoning.
Tool-driven workflows also fit when the hard part is deciding what to do next. Computer Use, Tool Search, Apply Patch, web and file search, and MCP allow GPT-5.4 Pro to connect deeper reasoning to actions rather than stopping at a recommendation.
- Difficult technical investigations with several interacting constraints.
- Long-context analysis that depends on relationships across many sources.
- High-value professional work where precision matters more than speed.
- Tool-driven workflows that require careful action selection.
- Escalation tiers for requests that lower-cost models solve unreliably.
- 01
Complex investigation
Reason across large evidence sets, competing explanations, and technical constraints before selecting a course of action.
- 02
Long-context synthesis
Keep many related sources in one working context and produce a coherent conclusion across them.
- 03
Computer-use workflows
Combine deeper reasoning with interface interaction when a task requires observation, action, and verification.
- 04
Escalated cases
Route only the hardest requests to Pro after a cheaper model or evaluation layer identifies insufficient confidence or quality.
03 / Pricing
GPT-5.4 Pro Pricing
GPT-5.4 Pro costs $30.00 per million input tokens and $180.00 per million output tokens, with no published cached-input discount.
Measure Cost Against the Value of a Better Answer
The price gap between GPT-5.4 and GPT-5.4 Pro is substantial. That makes cost-per-request a poor standalone metric for deciding whether Pro is justified.
The more useful metric is cost per successful outcome. If the model reduces retries, catches errors that would otherwise reach production, completes a difficult task without escalation to a human, or materially improves the reliability of a high-value workflow, the premium may be justified.
A request with 10,000 input tokens and 2,000 output tokens costs approximately $0.66 before any tool-specific charges. Large document sets, long reasoning traces, and verbose final outputs can increase total spend quickly.
OpenAI also applies a 10% uplift on regional processing endpoints for GPT-5.4 Pro.
- $30.00 per 1M input tokens.
- $180.00 per 1M output tokens.
- No published cached-input discount.
- Regional processing endpoints carry a 10% uplift.
- Tool-specific calls can add separate usage fees.
1M tokens · USD
- Input
- $30.00
- Cached input
- No discount
- Output
- $180.00
Example: 10K input + 2K output
- Input cost
- $0.3000
- Output cost
- $0.3600
- Estimated total
- $0.6600
04 / Context
A 1.05M-Token Window for High-Complexity Work
GPT-5.4 Pro provides a 1,050,000-token context window with up to 128,000 output tokens, allowing deep reasoning tasks to keep substantial technical, documentary, and conversational state together.
Large Context Is Most Valuable When the Connections Matter
The strongest reason to use a very large context window is not simply document size. It is the need to preserve relationships between many pieces of information during reasoning.
A technical investigation may require source code, architecture documentation, issue history, logs, earlier attempts, and tool results. A professional analysis may require multiple reports, policies, data extracts, and constraints that must all remain consistent.
GPT-5.4 Pro gives those workloads enough space to keep the relevant evidence available while the model reasons. But the absence of a cached-input discount makes repeated large-context requests expensive.
There is also a pricing threshold at 272K input tokens. Above that point, the entire request moves to higher long-context pricing. Context selection, retrieval quality, and unnecessary duplication therefore have a direct financial impact.
- 1,050,000-token context window.
- Maximum output of 128,000 tokens.
- Suitable for large technical and professional evidence sets.
- No cached-input discount for repeated context.
- Requests above 272K input tokens use higher pricing multipliers.
Context window
1,050,000
Max output
128,000
A Pro context may include source code, documents, policies, logs, prior reasoning state, tool results, and other evidence required to solve a difficult task.
05 / Reasoning
Medium to XHigh Reasoning
GPT-5.4 Pro supports medium, high, and xhigh reasoning effort, with medium as the default. Lightweight settings such as none and low are not available.
Every Request Starts with Deliberate Reasoning
The reasoning range communicates the intended role of the model. GPT-5.4 Pro does not offer a low-compute mode for routine work. Even its default setting is designed for tasks that benefit from a deliberate reasoning process.
medium provides the baseline Pro behavior. high can be tested when additional inference effort improves difficult planning, analysis, or tool selection. xhigh is the highest available reasoning level and should be reserved for cases where the incremental quality is worth the additional latency and compute.
This creates a natural routing strategy. If a task works reliably at none or low reasoning on another model, it is unlikely to need GPT-5.4 Pro. Pro becomes relevant when the workload already requires a deeper reasoning tier.
Some requests may run for several minutes, so applications should design long-running workflows around background execution rather than assuming interactive response times.
Pro baseline
Use medium to establish whether GPT-5.4 Pro improves the hard cases that justify moving beyond standard GPT-5.4.
Maximum reasoning
Test high or xhigh when the task has enough complexity and value to justify longer execution and additional compute.
06 / Capabilities
A Focused Agentic Toolset
GPT-5.4 Pro supports streaming, function calling, Computer Use, Apply Patch, Web Search, File Search, Image Generation, MCP, and Tool Search through the Responses API.
Deep Reasoning with Selective Execution Tools
GPT-5.4 Pro has a different capability profile from standard GPT-5.4. The Pro model is available through the Responses API and is designed around long-running, multi-turn reasoning before returning the final response.
Computer Use allows the model to interact with graphical interfaces. Apply Patch is useful for structured code modifications. Tool Search lets a large tool ecosystem defer definitions and expose relevant tools only when the model needs them.
Web Search and File Search support grounded reasoning over external information, while MCP connects the model to compatible external systems and tools.
There are important limitations. Structured Outputs are not supported. Code Interpreter, Hosted Shell, and Skills are also not supported for GPT-5.4 Pro. Fine-tuning is unavailable.
- Streaming is supported.
- Function calling is supported.
- Structured Outputs are not supported.
- Web Search and File Search are supported.
- Image Generation is supported.
- Apply Patch and Computer Use are supported.
- MCP and Tool Search are supported.
- Code Interpreter, Hosted Shell, and Skills are not supported.
- Fine-tuning is not supported.
- Supported
Computer use
Interact with graphical software interfaces during long-running agent workflows.
- Supported
Tool Search
Discover and load relevant tools from large tool ecosystems at runtime.
- Supported
Apply Patch
Apply structured code changes as part of supported technical workflows.
- Supported
Search and retrieval
Use web search and file search to ground reasoning in external information.
- Not listed
Structured Outputs
OpenAI currently lists Structured Outputs as unsupported for GPT-5.4 Pro.
- Not listed
Hosted execution
Code Interpreter and Hosted Shell are not supported for this Pro variant.
07 / Evaluation
Strengths and Limitations
GPT-5.4 Pro is a specialized model for difficult, high-value work: its deeper reasoning and agentic tools can improve hard outcomes, but premium cost, longer latency, and a narrower capability surface make selective routing essential.
Strengths
Higher-compute reasoning
GPT-5.4 Pro is explicitly designed to spend more compute on tough problems and produce smarter, more precise answers.
Large working context
A 1.05M-token window can keep complex evidence, project state, source material, and tool results together.
Computer-use workflows
Built-in computer use allows deep reasoning to continue into direct interaction with software interfaces.
Tool discovery and patching
Tool Search and Apply Patch make the model useful for agent systems that need to discover actions and modify code.
What to consider
Premium inference cost
At $30 input and $180 output per million tokens, GPT-5.4 Pro should be reserved for cases where additional quality has clear value.
Long-running requests
OpenAI notes that some requests may take several minutes and recommends background mode to avoid timeouts.
Responses API only
GPT-5.4 Pro is designed for the Responses API rather than as a general multi-endpoint model.
Reduced capability surface
Structured Outputs, Code Interpreter, Hosted Shell, Skills, and fine-tuning are not supported.
Evaluate the difficult cases
Test GPT-5.4 Pro where deeper reasoning could change the result
Run your hardest technical, analytical, long-context, and tool-driven tasks, then compare response quality, latency, token usage, and request cost in EidoStack.
Start FreeConnect your own OpenAI API key and evaluate GPT-5.4 Pro against the exact workloads that justify premium reasoning compute.
Common Questions
What is GPT-5.4 Pro?
GPT-5.4 Pro is a higher-compute version of GPT-5.4 designed for tough problems that benefit from deeper reasoning and more consistent answers.
How much does GPT-5.4 Pro cost?
GPT-5.4 Pro costs $30.00 per 1M input tokens and $180.00 per 1M output tokens. OpenAI does not publish a discounted cached-input rate for this model.
What is the context window of GPT-5.4 Pro?
GPT-5.4 Pro has a 1,050,000-token context window and supports up to 128,000 output tokens.
What reasoning levels does GPT-5.4 Pro support?
GPT-5.4 Pro supports medium, high, and xhigh reasoning effort. Medium is the default.
What is the knowledge cutoff for GPT-5.4 Pro?
OpenAI lists August 31, 2025 as the knowledge cutoff for GPT-5.4 Pro.
Does GPT-5.4 Pro support image input?
Yes. GPT-5.4 Pro accepts text and image input and produces text output. Direct audio and video modalities are not supported.
Does GPT-5.4 Pro support streaming?
Yes. OpenAI currently lists streaming as supported for GPT-5.4 Pro.
Does GPT-5.4 Pro support Structured Outputs?
No. OpenAI currently lists Structured Outputs as unsupported for GPT-5.4 Pro, although function calling is supported.
What tools does GPT-5.4 Pro support?
GPT-5.4 Pro supports Web Search, File Search, Image Generation, Apply Patch, Computer Use, MCP, and Tool Search. Code Interpreter, Hosted Shell, and Skills are not supported.
Why can GPT-5.4 Pro requests take longer?
GPT-5.4 Pro uses additional compute to reason through difficult problems. OpenAI notes that some requests may take several minutes and recommends background mode to avoid timeouts.
What workloads are a good fit for GPT-5.4 Pro?
GPT-5.4 Pro is best suited to complex technical investigations, high-value professional analysis, long-context synthesis, computer-use workflows, and escalation cases where deeper reasoning can justify premium cost.
Can GPT-5.4 Pro be fine-tuned?
No. The current OpenAI model documentation lists fine-tuning as unsupported for GPT-5.4 Pro.
Model information
Last updated
The specifications and prices on this page are based on the official OpenAI documentation for GPT-5.4 Pro. Provider pricing, tool support, reasoning controls, endpoints, model limits, regional processing, and availability may change, so production assumptions should be checked against the latest provider documentation.