OpenAI unveiled Daybreak, a cybersecurity platform combining GPT-5.5 and Codex to autonomously detect, validate, and patch software vulnerabilities. The launch escalates competition with Anthropic in the enterprise security AI market.
OpenAI doubled GPT-5.5 listed prices versus GPT-5.4, but OpenRouter analysis shows real user costs rose only 49-92% thanks to the model generating shorter completions.
OpenAI launched a limited preview of GPT-5.5-Cyber to vetted cybersecurity teams on May 7 via its Trusted Access for Cyber program — about a month after Anthropic's Mythos debut, despite OpenAI's earlier criticism of the restricted-access approach.
OpenAI launched GPT-5.5 Instant as ChatGPT's new default model, replacing GPT-5.3 Instant. The update delivers 52.5% fewer hallucinations on high-stakes topics like medicine, law, and finance, along with more concise responses and enhanced personalization using Gmail and past conversations.
DeepSeek V4 Pro tied with GPT-5.2 on FoodTruck Bench, a 30-day agentic benchmark using 34 tools, arriving roughly 10 weeks after GPT-5.2 was tested at approximately 17x lower cost.
The latest ARC-AGI-3 scores show GPT-5.5 High at 0.43% and Claude Opus 4.7 at 0.18% — the most powerful models today remain effectively at zero on this AGI benchmark.
The technique GPT-5.4 Pro used to solve Erdos Problem 1196 has been applied to other problems, including another conjecture unsolved for 60 years.
OpenAI announced GPT-5 on 2025-08-07 for both ChatGPT and API usage. The launch highlights include a reported 45% hallucination reduction vs GPT-4o and major benchmark gains such as HealthBench Hard 44.6.
OpenAI said on January 29, 2026 that ChatGPT would stop offering GPT-4o and older model options from February 13, 2026. GPT-4o, GPT-4.5, and o4-mini are being replaced by GPT-5, GPT-5 thinking, and o5-mini respectively.
OpenAI launches GPT-5.3-Codex, the first model to debug its own training and manage deployment. Released with tight security controls due to cybersecurity concerns.
A new study shows OpenAI's GPT-5 model outperformed federal judges in complex legal reasoning tasks.