OpenAI GPT 5.5 and 5.5 Pro launch with $25,000 bounty

OpenAI GPT 5.5 and 5.5 Pro have launched with enhanced intelligence and greater cybersecurity threats, requiring enhanced safeguards. Versus Claude Opus 4.7, GPT 5.5 / 5.5 Pro demonstrates better tool use, hacking skills, academic problem solving, and abstract reasoning.

OpenAI has released the latest versions of its AI models, GPT 5.5 and GPT 5.5 Pro, for its chatbot service, ChatGPT, and AI application programming interfaces (APIs). The new GPT models are smarter than GPT 5.4 and its competitors, Claude Opus 4.7 and Gemini 3.1 Pro, but they also come with a greater risk. GPT 5.5 and 5.5 Pro are available to paid ChatGPT users, and will be released soon for API use.

OpenAI’s new AIs continue to develop their capacity to solve complex academic questions and to use computers. The upgrades also offer a greater understanding of the protocols required to design biological weapons, and the skills required to effectively hack computer networks and systems. In many cases GPT 5.5 and GPT 5.5 Pro outperform other AI models Opus 4.7 and Gemini 3.1 Pro in these areas. Recently, Claude users have been seeing more vulnerable code being generated by the Anthropic AIs.

While extra precautions have been taken because of the high risk rating of OpenAI’s new models, OpenAI has launched a Bio Bug Bounty for GPT‑5.5 that will pay $25,000 to anyone who can jailbreak GPT 5.5 in Codex Desktop when presented with a five-question biosafety test. Readers can apply from April 23, 2026 to June 22, 2026.

Anthropic Claude Mythos is already so capable of discovering cybersecurity bugs that the company won’t release the AI for fear of the risk to US national security. Its less-powerful, public Claude Code has been able to hack FreeBSD.

Readers who want to run GPT locally can download and install the older, open-source GPT-OSS from Hugging Face, and run it on a PC with an Nvidia GPU of 16 GB or more memory (such as this 5090 GPU on Amazon).