The model is not intended to dramatically expand capabilities in the digital environment. Instead, it was trained for more flexible handling of security-specific workflows: vulnerability discovery and classification, patch verification, and malware analysis.

"GPT-5.5-Cyber will allow a limited group of partners to explore complex workflows where the specifics of access matter," the company stated in its blog.

The release comes one month after Anthropic unveiled Mythos — an AI model that drew widespread attention for its cybersecurity capabilities. Anthropic declined to release the model publicly due to significant safety risks, opting instead to launch Project Glasswing, an initiative for testing the tool under controlled conditions. OpenAI had previously responded in April with GPT-5.4-Cyber, focused on vulnerability discovery within its Trusted Access for Cyber programme launched in February.

Voice Models

OpenAI also announced three new real-time voice models:

  • GPT-Realtime-2 — the first voice model with reasoning capabilities on par with GPT-5. It handles more complex queries and maintains natural conversational flow, with five reasoning levels (minimal, low, medium, high, and very high) and a context window expanded from 32,000 to 128,000 tokens.
  • GPT-Realtime-Translate — a simultaneous translation model supporting live speech from more than 70 languages into 13, keeping pace with the speaker in real time. Context window: 16,000 tokens.
  • GPT-Realtime-Whisper — a real-time streaming speech-to-text system capable of generating subtitles during meetings, conferences, classes, and live streams, as well as producing notes and summaries during or immediately after a conversation.

Codex Extension for Chrome

OpenAI has released a Codex browser extension for Chrome on macOS and Windows. The tool interacts with applications and websites and runs in the background across all open tabs simultaneously.

"Codex can quickly handle routine browser tasks such as navigating structured pages and executing complex data entry processes," the announcement reads. At each step, the model independently selects the most appropriate tool for the task.

Emergency Contact in ChatGPT

OpenAI has added an Emergency Contact feature to ChatGPT. It allows a user — or a parent managing a child's account — to designate a trusted contact for crisis situations.

If the company's trained specialists detect that a user has been asking questions related to self-harm, the designated contact will receive a notification. The user can choose their preferred method of contact: a messaging app, email, or another ChatGPT account.