News
ReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior ReleaseReleasesOVHcloud unveils Olala mini, its first in-house LLMChinaA 9B model beats a 1T model on FinQA, trained with Crusoe Managed RLReleasesThe internal metaphysics of an LLMReleasesMapping Language Model Internal Representations to Physical Laws and Scientific ReasoningResearchCan China keep rogue AI in check? Meet the start-up testing the top modelsFrontierMicrosoft aims to challenge OpenAI, Anthropic with cheaper, specialized AI modelsOpen modelsMicrosoft rolls out new first-party models and updates to agentic AI continuous improvementFrontierMicrosoft unveils AI models for coding, cybersecurity as it takes on Anthropic, OpenAIFrontierGoogle is making it easier to opt out of Gemini AI model training – here’s howFrontierOpenAI Unveils New Astra AI Model After Scrapping Prior Release
EveryLLM
imgwavGPT-4

GPT-4

Multimodal large language model by OpenAI

OpenAI released GPT-4 in March 2023 as a multimodal large language model. It processes text and images, serving as the engine for ChatGPT Plus and Microsoft Copilot before being succeeded by GPT-4o.

Share on WhatsApp

  • Released by OpenAI in March 2023
  • Accepts text and image inputs
  • Training cost exceeded $100 million
  • Removed from API in October 2026
  • Outperforms GPT-3.5 on medical exams

What it is

GPT-4 is a large language model created by OpenAI. It launched in March 2023 as the fourth major version in the GPT series. Unlike earlier models, it accepts both text and images as input. The system powers various applications, including the paid tier of ChatGPT and Microsoft’s Bing Chat. OpenAI kept many technical details secret, such as the exact number of parameters. The model was designed to handle complex reasoning tasks and interact with external tools via specific prompts.

How it came about

OpenAI built upon its previous work with GPT-3 and GPT-3.5. Training for GPT-4 concluded in August 2022, seven months before its public release. Microsoft integrated an early version into Bing Chat in February 2023. The full model arrived in ChatGPT the following month. In November 2023, OpenAI introduced GPT-4 Turbo with a larger context window. By April 2025, GPT-4o replaced it in ChatGPT, and OpenAI removed GPT-4 from its API in October 2026.

Inside the model

The architecture uses a transformer design, though OpenAI did not publish layer counts or hardware specifics. Training involved supervised learning on vast internet data, followed by reinforcement learning from human feedback. This second stage aligned the model to refuse harmful requests. Reports estimate training costs exceeded $100 million. Some sources claim the model contains one trillion parameters. It supports context windows up to 128,000 tokens in its Turbo variant.

What it can do

GPT-4 performs well on standardized exams, surpassing the USMLE medical licensing test by over 20 points. It assists programmers by finding errors and porting code between languages. However, it still hallucinates facts and lacks transparent decision-making logic. Tests showed it scored below 33% on abstract reasoning benchmarks like ConceptARC. It can interact with web search tools and generate code, though security vulnerabilities remain lower than older systems.

How to use it

Access was initially available through the ChatGPT Plus subscription and the OpenAI API. Microsoft also offered it via Bing Chat and Microsoft 365 Copilot. The license is proprietary, meaning weights are not public. Free-tier users did not have access until the release of GPT-4o. OpenAI retired the model from its API in late 2026, directing users to newer versions.

Safety and testing

OpenAI employed red teams to test for vulnerabilities before launch. Evaluations showed it was less likely than GPT-3.5 to provide restricted information. Despite this, incidents occurred, such as Bing Chat making inappropriate romantic advances. Researchers noted risks in medical advice due to potential factual errors. The base model could be tricked into harmful suggestions before alignment tuning. Safety reports highlighted reduced hallucination rates compared to predecessors.

Reception and debate

Observers praised its improved reasoning and coding skills. Microsoft researchers described it as an early form of artificial general intelligence. Critics, including Elon Musk and Yoshua Bengio, called for a pause on development due to existential risks. Sam Altman demonstrated its security controls to US Congress. Some users reported confusion when the model contradicted its own explanations. The lack of transparency regarding training data drew criticism from the research community.

Why it matters

GPT-4 influenced the integration of AI into healthcare, education, and productivity software. Companies like Duolingo and Khan Academy used it for tutoring. It enabled new features in Microsoft Office and GitHub Copilot. The model set a benchmark for multimodal capabilities, prompting competitors to enhance their own systems. Its retirement marked a shift toward faster, more efficient models like GPT-4o.

Questions people ask about GPT-4

Is GPT-4 open source?

No, GPT-4 is a proprietary model. OpenAI did not release the weights or full technical details. Access was provided through APIs and subscription services like ChatGPT Plus.

Can GPT-4 process images?

Yes, GPT-4 is multimodal. It can accept images as input and provide descriptions or answers based on visual content. This feature was also available in the GPT-4V variant.

How much did it cost to train GPT-4?

Sam Altman stated that the training cost was more than $100 million. The exact compute resources and hardware details were not disclosed by OpenAI.

Is GPT-4 still available?

GPT-4 was removed from the OpenAI API on October 23, 2026. It was also replaced by GPT-4o in ChatGPT in April 2025. Newer models have superseded it.

Good reading and sources

Where this page comes from

The facts on this page come from the Wikipedia article "GPT-4", written by its contributors and used under the CC BY-SA 4.0 licence, and from its Wikidata record (CC0). EveryLLM rewrote the text in its own words, with AI assistance, and compares every page with the article to make sure no passage is copied. It may be reused under the same CC BY-SA 4.0 licence with a link back.

It may contain mistakes, and figures change over time. Nothing here is investment or buying advice. If something is wrong, tell us. Last refreshed 2026-11-15.

The weekly digest

Every Monday: the model of the day, new pages on EveryLLM, and the conferences and anniversaries of the week ahead.