GPT-5.2 Pro

What is GPT-5.2 Pro?

GPT-5.2 Pro is a professional reasoning model from OpenAI. It is intended for problems that benefit from additional inference computation, such as evaluating complicated evidence, solving demanding technical problems, reviewing large documents, writing or debugging code, and performing multi-step analysis.

The “Pro” designation indicates a trade-off rather than a universal upgrade for every task. GPT-5.2 Pro can spend more time reasoning through a request, which may improve results on difficult work, but it is more expensive and can be slower than general-purpose models. Some requests may take several minutes, so applications should account for long-running responses instead of assuming that every request will complete quickly.

OpenAI introduced GPT-5.2 Pro on December 11, 2025. It belongs to the GPT-5.2 model generation and should be distinguished from both the standard GPT-5.2 API model and the consumer ChatGPT experience that used GPT-5.2 models.

Current status and availability

OpenAI's current documentation describes GPT-5.2 Pro as a previous Pro model and recommends GPT-5.5 Pro for new deployments. However, the exact gpt-5.2-pro model remains available through the Responses API according to the supplied model documentation.

The model is restricted to the Responses API for direct model requests. It is not available through Chat Completions. This distinction matters when planning an integration: an application must use the Responses API interface and should not assume that code written for a Chat Completions request can be transferred unchanged.

GPT-5.2 models were removed from ChatGPT on June 12, 2026. That consumer-product change is separate from API availability. It does not establish that the GPT-5.2 Pro API model has been shut down.

GPT-5.2 Pro specifications

The 400,000-token context window is the amount of input and conversation material the model can handle within a request context, subject to the API's current operational rules. It is useful for lengthy reports, codebases, collections of documents, and multi-step workflows. The maximum output is 128,000 tokens, although producing such a large response is not appropriate for most requests and may increase cost and latency.

Inputs, outputs, and modalities

GPT-5.2 Pro accepts both text and image input. Image understanding can support tasks such as interpreting diagrams, examining screenshots, reviewing scanned pages, and extracting meaning from visual documents. The model's output is text.

This is multimodal input, not multimodal generation. GPT-5.2 Pro does not natively produce images, audio, or video. A workflow that needs generated artwork, speech, music, or video requires a separate compatible tool or model. The model can still participate in such a workflow through function calling if an application connects it to external services, but that does not make those output formats native capabilities of GPT-5.2 Pro.

Reasoning and coding capabilities

GPT-5.2 Pro is designed for demanding reasoning rather than only rapid conversational responses. Its reasoning effort can be set to medium, high, or xhigh. Higher settings give the model more scope to work through a difficult problem, but they can also increase processing time and token usage. A practical approach is to use medium for routine analysis and reserve high or xhigh for problems where additional quality justifies the delay and expense.

Suitable reasoning tasks include comparing competing explanations, synthesizing research, working through advanced mathematics, evaluating scientific information, and making a structured assessment from a large supplied context. These are capability areas supported by the model's positioning and documented reasoning controls; they should not be treated as guarantees of factual accuracy.

For programming, GPT-5.2 Pro is appropriate for substantial code generation, debugging, architecture review, refactoring, and codebase analysis. Its long context can help when a task involves many related files or extensive technical documentation. However, generated code still requires testing, security review, and validation against the target environment. The model's knowledge cutoff is August 31, 2025, so current libraries, APIs, or platform behavior should be checked using supplied documentation or external tools.

Tools and API features

GPT-5.2 Pro supports function calling. Function calling allows an application to define operations that the model may request, such as querying a database, retrieving a document, invoking a calculator, or sending work to another service. The application, rather than the model alone, performs the operation and returns the result for further reasoning.

Streaming is also supported, allowing an application to receive response content incrementally instead of waiting for the entire response. This can improve the user experience for long answers, although streaming does not necessarily reduce the total processing time.

The model is documented as not supporting structured outputs or fine-tuning. Structured outputs are stricter than simply asking the model to write JSON: they provide a mechanism for enforcing a specified response schema. GPT-5.2 Pro should therefore not be selected for an integration that depends on provider-enforced structured responses. Likewise, organizations cannot use the documented fine-tuning capability to create a specialized version of this model.

The supplied research does not verify native web search support for GPT-5.2 Pro. Applications that need current information should use an explicitly connected search or retrieval tool and treat retrieved material as external context.

GPT-5.2 Pro pricing

GPT-5.2 Pro costs $21 per 1 million input tokens and $168 per 1 million output tokens. The supplied model information does not list a separate cached-input price.

Input tokens include the material sent to the model, while output tokens represent the generated response. Long prompts, large documents, high reasoning effort, and lengthy answers can all affect consumption. The output price is eight times the input price, so applications should avoid requesting unnecessarily verbose responses when a concise result is sufficient.

Batch processing is exposed through OpenAI's Batch API infrastructure, but applicable batch pricing and queue limits are determined by the current Batch API rules. GPT-5.2 Pro is therefore best suited to high-value tasks where improved reasoning quality can justify higher inference cost and potentially longer waits, rather than inexpensive, high-volume classification or simple conversational traffic.

Main strengths and limitations

Strengths

• Complex reasoning: configurable reasoning effort is aimed at difficult analytical, scientific, mathematical, and professional tasks.
• Large context: the 400,000-token context window supports long documents, extensive code, and multi-step work.
• Visual analysis: image input extends the model beyond text-only analysis.
• Professional coding: the model is suitable for substantial code generation, debugging, review, and technical explanation.
• Tool integration: function calling supports applications that need external data or actions.
• Long responses: the documented maximum output is 128,000 tokens, subject to practical cost and latency considerations.

Limitations

• Higher cost: $21 per million input tokens and $168 per million output tokens make it substantially more expensive than ordinary general-purpose options.
• Lower speed: additional reasoning can produce latency of several minutes for some requests.
• API restriction: direct access is through the Responses API, not Chat Completions.
• Text-only output: it does not natively generate images, audio, or video.
• No documented structured outputs: applications requiring schema-enforced responses should consider another option.
• No fine-tuning: the supplied documentation lists fine-tuning as unsupported.
• Knowledge cutoff: the model's built-in knowledge ends on August 31, 2025, so current facts need external information.
• Previous-generation status: OpenAI recommends GPT-5.5 Pro for new deployments, so teams should evaluate whether the newer model is a better long-term choice.

Best use cases

GPT-5.2 Pro is a strong fit when the cost of an incorrect or shallow answer is higher than the cost of additional computation. Examples include:

• Research synthesis across lengthy reports and source material
• Advanced mathematics and scientific reasoning
• Complex programming, debugging, and code review
• Analysis of large technical or legal-style document sets
• Agentic workflows that combine reasoning with function calls
• Professional analysis where quality is more important than immediate response time

For example, an application could supply several technical reports, ask the model to identify conflicting findings, and then use a function call to retrieve additional evidence. A software team could provide a large set of related source files and ask for a dependency-aware refactoring plan. In both cases, the long context and reasoning controls are more relevant than fast conversational turn-taking.

When should you choose GPT-5.2 Pro?

Choose GPT-5.2 Pro when you need a high-end reasoning model that can work with large inputs, analyze images, call external functions, and produce text through the Responses API. It is especially appropriate when a difficult task may require several reasoning steps and a slower, more expensive response is acceptable.

Choose a faster or less expensive general-purpose model when the task is routine summarization, simple extraction, basic drafting, short answers, or high-volume automation. Paying Pro-level rates for those tasks may not produce enough additional value to justify the cost.

Choose a model or workflow with native image, audio, or video generation when those formats are the required outputs. Choose a structured-output-capable model when downstream software depends on strict schema conformance. For new OpenAI deployments, compare GPT-5.2 Pro with GPT-5.5 Pro, which OpenAI recommends as the newer Pro option. The existing GPT-5.2 model may also be a more suitable comparison when the task does not require the Pro model's additional reasoning focus.

Implementation guidance

Use the canonical model ID gpt-5.2-pro with the Responses API. Select reasoning effort according to task value: medium for a balance between quality and latency, high for more demanding analysis, and xhigh when the additional processing is justified. For requests that may take several minutes, use background processing where appropriate and design the client to tolerate delayed completion.

Keep model limitations visible in the application design. Validate generated code, verify important factual claims, and provide current documents or retrieval results when the knowledge cutoff is insufficient. Also separate consumer ChatGPT availability from API availability: the retirement of GPT-5.2 models from ChatGPT does not by itself mean that the GPT-5.2 Pro API model is unavailable.