GPT-5
Multimodal large language model released by OpenAI in August 2025.
OpenAI launched GPT-5 on August 7, 2025. It is a multimodal system that integrates fast and reasoning models to handle text, images, and code tasks.
- Released on August 7, 2025 by OpenAI.
- Combines fast and reasoning models with a router.
- Trained natively on text and images from scratch.
- Available in ChatGPT, Copilot, and via API.
- Critics noted inconsistent quality due to auto-switching.
What it is
GPT-5 is a multimodal large language model developed by OpenAI. It serves as the fifth iteration in the company's generative pre-trained transformer series. The model was officially released on August 7, 2025. It functions as a foundation model capable of processing multiple data types simultaneously. Users access it through ChatGPT, Microsoft Copilot, and the OpenAI API. The system replaces earlier models like GPT-4.1 and OpenAI o3 in standard offerings.
How it came about
OpenAI initially stated in April 2023 that it was not training GPT-5. The company filed for a trademark in July 2024. An internal project named Orion failed to produce superior results and became GPT-4.5. Development resumed for GPT-5 later in 2024. Sam Altman confirmed work on the model in November 2024. Anticipation grew in July 2025 as Microsoft prepared its Copilot integration. OpenAI released open-weight reasoning models before the main launch. A livestream event unveiled the final product on August 7, 2025.
Inside the model
The architecture combines a fast high-throughput model with a deeper reasoning model. A real-time router selects the appropriate model based on task complexity and user intent. Training involved unsupervised pretraining, supervised fine-tuning, and reinforcement learning from human feedback. The model was trained natively on multiple modalities from scratch. It includes agentic functionality for autonomous browser use and desktop setup. Developers can adjust reasoning effort and verbosity levels via the API.
What it can do
GPT-5 demonstrated strong performance in mathematics, programming, finance, and multimodal understanding. It offers faster response times and improved coding skills compared to predecessors. The model provides more accurate health information and reduces hallucinations. It aims to give safe responses to harmful queries rather than outright refusals. Testers noted significant improvements in solving scientific problems. Some observers described its capabilities as reaching PhD-level proficiency across various domains.
How to use it
The model is available through the OpenAI API for developers. Public users access it via ChatGPT and Microsoft Copilot. ChatGPT offers different tiers: free users have standard limits, Plus users get higher limits, and Pro users receive unlimited access. The API allows selection between main and thinking variants. Thinking models support adjustable compute effort. Microsoft plans to integrate the model into various products. Apple also intends to include it in future operating systems.
Safety and testing
Security researchers claimed to compromise the model within its first day. Reports indicated it could generate instructions for manufacturing explosives. Similar assessments by other firms suggested significant security gaps. These findings raised concerns about corporate safety. OpenAI implemented safe completion strategies to handle harmful queries better. The goal was to refuse unsafe questions while providing harmless information. Researchers estimated energy consumption for medium responses at approximately 18 watt-hours.
Reception and debate
Reviews highlighted refined user experience over transformative breakthroughs. Critics noted the model felt intuitive and efficient but lacked the excitement of earlier leaps. Users criticized the automatic router for inconsistent quality and occasional performance drops. Some found the tone flat and uncreative compared to previous versions. Altman acknowledged underestimating user preference for warmer tones. OpenAI rolled out updates to improve personality warmth. Hype surrounding the release led to criticism of unrealistic expectations set by leadership.
Why it matters
The model influenced how companies deploy AI assistants. Uber used it for customer support, while GitLab adopted it for development. Financial institutions applied it for analysis. The integration into major platforms expanded its reach. The router architecture introduced new paradigms for model selection. Feedback on tone and control drove changes in customization features. The release marked a shift toward more seamless and integrated AI interactions.
Questions people ask about GPT-5
What is GPT-5?
GPT-5 is a multimodal large language model developed by OpenAI. It is the fifth generation in the GPT series. The model processes text, images, and code. It was released on August 7, 2025. Users access it through ChatGPT and the API.
How does GPT-5 work internally?
It uses a system with a fast model and a deeper reasoning model. A router decides which model to use based on the task. This allows it to balance speed and depth. Developers can adjust reasoning effort levels. The architecture supports agentic functions like browser use.
Where can I use GPT-5?
You can use it in ChatGPT and Microsoft Copilot. Developers can access it through the OpenAI API. Different subscription tiers offer varying usage limits. Free users have standard restrictions. Pro users get unlimited access to certain variants.
What are the main criticisms?
Users reported inconsistent quality from the auto-router. Some found the tone too flat compared to older models. Security researchers claimed it could be compromised. Critics said it offered refinement rather than major breakthroughs. Leadership faced backlash for overhyping capabilities.
Good reading and sources
- Official websiteOfficial site
- Microsoft brings GPT-5 to Copilot with new smart modeThe Verge, Article
- OpenAI claims new GPT-5 model boosts ChatGPT to ‘PhD level’BBC, Article
- OpenAI Finally Launched GPT-5. Here's Everything You Need to KnowWired, Article
The facts on this page come from the Wikipedia article "GPT-5", 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 2026-11-17.