OpenAI’s September 3 launch of GPT-6 Astra, its latest frontier large language model, supplies the main catalyst for trader sentiment on Humanity’s Last Exam (HLE) debut performance. Official benchmarks released with the model show Astra scoring 57.2% on HLE with tools—below the 65% posted by its predecessor GPT-5.6 Sol and several competing systems—while leading on computer-use, browser, coding, and cybersecurity tasks such as ExploitBench and Terminal-Bench. Astra’s rollout followed weeks of delays triggered by its “critical” cybersecurity classification under OpenAI’s Preparedness Framework, limiting advanced exploit capabilities at launch. Full availability to ChatGPT paid users and the API begins in the coming days, with potential for tool or agent updates to influence final leaderboard entries.