LLM

GPT-4

Review GPT-4 retirement dates, historical API pricing, text versus image-capable variants, context limits and practical checks for legacy projects.

Updated October 29, 2026

Availability update: OpenAI now marks GPT-4 as shut down. Its API deprecations page lists October 23, 2026 for gpt-4-0613 and the gpt-4 alias, as well as GPT-4 Turbo. GPT-4-32k had already shut down on June 6, 2025. Do not treat an older pricing table or launch-page access button as proof of current availability.

Image context: The featured image is an earlier OpenAI GPT-4 webpage screenshot, retained for historical product identification. Its original capture date is unknown; it was reviewed on October 29, 2026. It is not a current interface, generated result or hands-on test. OpenAI branding belongs to OpenAI; this independent article does not imply endorsement. Source: March 2023 research announcement.

Name	GPT-4
Overview	OpenAI introduced GPT-4 in March 2023 as a language-model research milestone. This page is now a historical and technical reference, not an offer of access to the retired model.
Key features & benefits	Text prompting and text responses for language tasks. Instruction-following and adaptable conversational applications. Image understanding in specific multimodal variants, not every GPT-4 API model. Published benchmark results that need to be read in their original evaluation context.
Use cases and applications	Historical applications included summarizing documents, drafting content, studying language-model behavior and building conversational systems. Image captioning or interpretation required an image-capable variant. These examples are not recommendations to start a new integration with a shut-down model.
Who uses?	Researchers, data scientists, AI professionals and students studying earlier models or maintaining records of legacy integrations.
Pricing	There is no current GPT-4 freemium offer established here. Historical API rates were $0.03 per 1,000 input tokens and $0.06 per 1,000 output tokens for the 8K model; the separate 32K variant was $0.06 and $0.12 respectively. These were usage-based rates, not two subscription plans or current purchase offers.
Tags	AI, language models, deep learning, natural language processing, multimodal research, legacy API models
App available?	GPT-4 was a model accessed through products and an API, not a standalone native app. API integration does not establish an app download. Its former presence in ChatGPT is distinct from API model availability.

GPT-4, image inputs and context limits

The GPT-4 model documentation describes a text-input/text-output model with an 8,192-token context window and no image support. By contrast, the now-retired GPT-4 Turbo documentation lists image input, text output, a 128,000-token context window and a 4,096-token maximum output. A context window is not a promise that the model can produce an answer of that length.

The March 2023 research announcement described a multimodal research model and reported strong results on selected professional and academic benchmarks. Those were OpenAI’s historical evaluations, not a guarantee of human-level judgment, factual accuracy or performance on your task. The announcement initially made text access available while image access was limited.

Historical pricing and ChatGPT access

The 8K and 32K rates above come from the launch announcement. The GPT-4 model reference still displays $30 input and $60 output per million tokens, equivalent to the old 8K rates; a retained price does not reverse its shutdown status. The ChatGPT release notes separately record GPT-4’s retirement from ChatGPT on April 30, 2025. Its statement at that time that API access would continue should not override the later API shutdown notice.

A practical review example

For a hypothetical legacy document-summary project, record the exact model ID used in old logs, whether the task required text or images, and the input length. Compare an archived summary with its source: check names, dates, missing qualifications and invented statements. If rebuilding the project, choose a currently supported model from official documentation and evaluate it separately rather than assuming an old GPT-4 result transfers unchanged. This is a suggested review process, not a benchmark or an API test performed for this article.

Limits and data handling

Review generated text before relying on or publishing it. A historical model description does not establish the privacy settings, retention rules or permissions of a present-day app using AI. Check the terms and data controls of the actual service you intend to use, and do not submit confidential documents without appropriate authorization.