Short answer: For GPT-4.1 API specifications, use OpenAI’s model reference rather than third-party model-card listings. As of November 12, 2026, that reference labels GPT-4.1 deprecated and lists a 1,047,576-token context window, 32,768 maximum output tokens, text and image input, and text output. It lists audio and video as unsupported. The reference is an API specification, not a complete safety or limitations model card. (OpenAI API model reference)
For a new production service, don’t assume GPT-4.1 is the right default: check its deprecated status, confirm availability for your API project, and evaluate a currently supported candidate against your workload. If you already use GPT-4.1, plan any migration using representative tests rather than relying on broad capability claims.
What OpenAI’s API reference establishes
The table summarizes the OpenAI model reference as checked on November 12, 2026. Status and pricing can change; recheck the linked documentation before making a deployment or budget decision.
| Item | OpenAI API reference |
|---|---|
| Model ID | gpt-4.1 |
| Status | Deprecated |
| Context window | 1,047,576 tokens |
| Maximum output | 32,768 tokens |
| Knowledge cutoff | June 1, 2024 |
| Input and output | Text and image input; text output |
| Audio and video | Not supported |
| Supported endpoints listed | Chat Completions and Responses |
| Listed text-token rates | $2 input, $0.50 cached input and $8 output per 1 million tokens |
The model reference displays prices with dollar signs and a per-million-token unit. Confirm the applicable pricing category, billing terms and currency with the current pricing documentation and your account before estimating spend. It also says rate limits depend on usage tier; check the limits for your project. Confirm that the endpoint and features your application requires are documented for your intended use. (OpenAI API model reference)
Why third-party context figures differ
The third-party Simtheory page lists a 104,757-token context window; a CometAPI article reports 1 million tokens. Those figures conflict and the captured third-party material does not substantiate them with an official specification. OpenAI’s API reference lists 1,047,576 tokens for the documented model. It also lists a June 1, 2024 knowledge cutoff; Simtheory’s different cutoff and displayed sampling parameters should not be presented as official API defaults. (Simtheory · CometAPI · OpenAI API model reference)
Capabilities: useful for choosing what to test, not proof of fit
OpenAI’s April 14, 2025 launch announcement introduced GPT-4.1, GPT-4.1 mini and GPT-4.1 nano for the API, and described gains in coding, instruction following and long-context understanding. These are launch-era positioning and evaluation claims; they do not establish how the model performs on your data, prompts or production constraints. Test your own representative tasks before choosing a model. (OpenAI launch announcement)
Keep modality claims specific: the API reference lists image input and text output, and marks audio and video unsupported. A third-party description that broadly mentions audio is not evidence of audio support for this API model. (OpenAI API model reference)
Production API checks before you commit
- Status and access: The model reference labels GPT-4.1 deprecated. Confirm that the model ID is available to your API project; the public reference does not verify access for your account.
- Inputs and outputs: Check that text or image input and text output meet your requirements. If your application needs audio or video, the reference marks those modalities unsupported.
- Endpoint and features: The reference lists Chat Completions and Responses. Confirm that your required features are documented for the endpoint you plan to use, and check its request requirements.
- Limits and cost: Check your project’s rate limits and calculate expected input, cached-input and output token use against the current listed rates. Confirm the applicable pricing category and billing terms before budgeting.
- Lifecycle: OpenAI announced GPT-4.1’s retirement from ChatGPT on February 13, 2026, while saying there were no API changes at that time. That ChatGPT change is not an API shutdown date. Separately, the API model reference now marks GPT-4.1 deprecated; the model details reviewed here do not state an API shutdown date. (OpenAI retirement announcement · OpenAI API model reference)
Illustrative token-cost calculation
Using the listed dollar rates, a hypothetical request with 100,000 uncached input tokens and 2,000 output tokens would calculate as:
- Input: 100,000 ÷ 1,000,000 × $2 = $0.20
- Output: 2,000 ÷ 1,000,000 × $8 = $0.016
- Total: $0.216, before any other applicable charges or usage conditions
This is arithmetic based on the displayed per-million-token rates, not a measured API bill. Confirm the rates and pricing category that apply to your account before using the estimate.
Should you choose GPT-4.1?
- For a new integration: Since the API reference labels GPT-4.1 deprecated, first identify a currently supported candidate and compare it against your task, quality threshold, latency needs, input patterns and budget.
- For an existing GPT-4.1 service: Confirm project access and current limits. Establish a baseline, test a replacement on the same representative requests—including important edge cases—and compare quality, latency and cost before changing production traffic. OpenAI’s migration guidance identifies GPT-6 Luna as a starting point for existing GPT-4.x use cases; treat it as a candidate to evaluate, not an automatic replacement. (OpenAI migration guidance)
- For image, audio or video workloads: Treat modality support as a selection requirement. The API reference lists image input, text output, and no audio or video support for GPT-4.1.
Bottom line: OpenAI’s API model reference resolves the conflicting third-party context-window figures and documents GPT-4.1’s limits, modality support, Chat Completions and Responses endpoints, and token rates. It also marks the model deprecated. Use the dated reference as a specification snapshot, verify access and terms for your project, and base a production decision on representative workload testing—not headline claims alone.
Sources
Introducing GPT-4.1 in the API
Retiring GPT-4o, GPT-4.1, GPT-4.1 mini, and OpenAI o4-mini in ChatGPT
Migrating from GPT-4.x to GPT-6 Luna