OpenAI has launched GPT-5.5-Cyber for cybersecurity tasks and introduced three new voice models, expanding its AI capabilities.
OpenAI has unveiled GPT-5.5-Cyber, a specialized version of its AI model tailored for cybersecurity applications. Currently, this model is accessible to a limited group of developers.
Rather than expanding general capabilities, GPT-5.5-Cyber focuses on enhancing the flexibility of security-related tasks, including vulnerability identification, patch verification, and malware analysis.
According to OpenAI, “GPT-5.5-Cyber will allow a select group of partners to explore complex workflows where access specificity matters.”
This launch follows the introduction of Mythos, a competing AI model from Anthropic, which is known for its cybersecurity features but has not been made publicly available due to security concerns. Instead, Anthropic is testing its capabilities through Project Glasswing.
In response to the competitive landscape, OpenAI also released GPT-5.4-Cyber in April, aimed at identifying vulnerabilities as part of its Trusted Access for Cyber initiative launched earlier this year.
New Voice Models Introduced
In addition to the cybersecurity model, OpenAI has introduced three innovative voice models:
- GPT-Realtime-2: This model possesses GPT-5-level reasoning skills, enabling it to handle complex inquiries and engage in natural dialogue.
- GPT-Realtime-Translate: A language model designed for real-time translation, capable of keeping pace with spoken language.
- GPT-Realtime-Whisper: A streaming speech-to-text system that functions in real time, providing subtitles and notes during discussions.
OpenAI shared on Twitter, “Introducing GPT-Realtime-2 in the API: our most intelligent voice model yet, bringing GPT-5-class reasoning to voice agents.”
GPT-Realtime-2 features five reasoning levels and has significantly expanded its context window from 32,000 to 128,000 tokens. The translation model supports over 70 languages, translating into 13 languages with a context window of 16,000 tokens. Meanwhile, GPT-Realtime-Whisper can generate real-time subtitles and summaries.
These voice models are now available through the Realtime API.
New Chrome Extension and Safety Features
OpenAI has also introduced a Codex extension for Chrome, compatible with both macOS and Windows. This extension operates in the background across multiple tabs, facilitating tasks like navigating web pages and executing data entry efficiently.
OpenAI noted, “Codex can quickly handle routine tasks in the browser.”
Additionally, OpenAI has implemented an “Emergency Contact” feature in ChatGPT, enabling users, or parents of child accounts, to designate a contact for emergencies. If the system detects concerning inquiries regarding self-harm, it will notify the designated contact through the selected communication method.
This feature builds on the parental controls that OpenAI introduced in September 2025.
