In a surprising development, OpenAI‘s next-generation AI system has been accidentally revealed to the public. The company is testing a specialized version of GPT-5 called “GPT-Alpha,” which briefly appeared in ChatGPT’s model selector before quickly disappearing. This leak confirms that OpenAI is developing an advanced AI agent rather than just another language model.
The leaked feature, called “Agent with Truncation,” showed that GPT-Alpha can perform several impressive tasks. It can browse the web to find current and niche information beyond its training data. The system can generate and edit images directly. It can also write, run, and debug code to handle complex programming tasks. Additionally, it creates and edits documents, including spreadsheets and slides.
GPT-5 shows major improvements over earlier models. It reasons better and thinks more efficiently, using 50-80% fewer output tokens for similar tasks. The model is more honest about its limitations and less likely to make confident but wrong statements. It combines visual reasoning with language understanding. OpenAI trained it using Microsoft Azure AI supercomputers. The testing is currently internal within OpenAI with no public release date confirmed.
GPT-5 elevates reasoning, efficiency, and honesty, using dramatically fewer tokens while integrating visual understanding with supercomputing power.
The company plans to release four versions of GPT-5. The full version offers maximum capabilities. GPT-5-mini uses fewer resources for mobile devices. GPT-5-nano focuses on speed and low latency. GPT-5-chat is designed for conversation and multi-modal interactions. Each variant serves different needs and hardware limitations. Similar to other multimodal AI systems, it can process text, images, and audio in an integrated way that enhances user experience.
Signs of GPT-5’s development appeared earlier this year. AI models with codenames like “Zenith,” “Summit,” and “Nectarine” showed up on benchmark sites. Brief API endpoints and hidden model previews appeared online before being removed. Google briefly indexed GPT-5 documentation pages. Users on Reddit reported seeing models with better coding skills than GPT-4o.
OpenAI will likely offer these advanced agent features to paying customers due to their higher computing demands. The company has included privacy safeguards to prevent the system from accessing private data without permission. The appearance and quick removal of these agents suggests an accidental public exposure of what was meant to be confidential internal testing.
References
- https://www.bleepingcomputer.com/news/artificial-intelligence/openai-is-testing-a-new-gpt-5-based-ai-agent-gpt-alpha/
- https://sqmagazine.co.uk/openai-gpt5-alpha-agent-test/
- https://autogpt.net/github-leak-reveals-openai-gpt-5/
- https://www.revolgy.com/insights/blog/preparing-for-gpt-5-what-we-know-what-to-expect-and-whats-rumored
- https://openai.com/index/introducing-gpt-5/
- https://openai.com/gpt-5/
