OpenAI’s GPT-6 Astra Draws Rough Safety Comparison With Gemini Flash and Claude Opus 4.8

OpenAI’s GPT-6 Astra has been described by Anthropic’s Boris Cherny as roughly comparable with Gemini Flash and Claude Opus 4.8 on prompt-injection risk, while OpenAI’s own testing reports a substantial improvement over GPT-5.6 Sol. The comparison concerns an attacker’s ability to manipulate a model through hostile instructions embedded in external content, a central risk for browsing and tool-using systems.