- Will OpenAI’s Astra model debut with a Humanity’s Last Exam score of 40% or higher?98¢
- Will OpenAI’s Astra model debut with a Humanity’s Last Exam score of 50% or higher?65¢
- Will OpenAI’s Astra model debut with a Humanity’s Last Exam score of 60% or higher?6¢
On August 1, 2026, OpenAI publicly disclosed “Astra,” describing an internal version of Astra as its “next major model.” You can read more about that here: https://openai.com/index/ten-advances-in-mathematics/. This market will resolve to "Yes" if the next OpenAI “Astra” model added to the Humanity’s Last Exam results at https://agi.safe.ai/ has an HLE Accuracy of at least the specified percentage at 12:00 PM ET on the calendar date following the date on which it first appears on the site. Otherwise, this market will resolve to "No". If a model first appears on the site but is removed before 12:00 PM ET on the following calendar date, its appearance will not qualify as added to the Humanity’s Last Exam results. A qualifying model must be named "Astra" (including variants or successor branding such as Astra 1, Astra 6, Astra X, GPT-6 Astra, or similar naming or rebranded versions), or be confirmed to be the same model referenced in the announcement by OpenAI or by a consensus of credible reporting. The percentage displayed as “HLE Accuracy” for the model in its result card on the “AI Progress on Humanity’s Last Exam” chart at https://agi.safe.ai/ will be used to resolve this market. This market will resolve solely based on the displayed HLE Accuracy, regardless of the model’s Calibration Error, chart position, other configuration scores, or any underlying granular or unrounded data presented elsewhere. If multiple qualifying models are added to the Humanity’s Last Exam results on the same calendar date (ET), the model with the highest HLE Accuracy will be used for resolution. Models added to the results on the calendar date following the initial qualifying model’s first appearance will not be considered. A qualifying model must be newly added to the Humanity’s Last Exam results at https://agi.safe.ai/. Whether the model was previously released, publicly accessible, in beta, or otherwise available before appearing on the site is irrelevant for this market. The resolution source for this market is the official Humanity’s Last Exam website found at https://agi.safe.ai/. If this resolution source is unavailable at 12:00 PM ET on the calendar date following the date on which the qualifying model first appears on the site, this market will resolve based on the first subsequent instance at which the model’s HLE Accuracy becomes available on the site. If it remains unavailable through the end of the seventh day after the qualifying model first appears on the site or if no qualifying model is added by December 31, 2026, 11:59 PM ET, this market will resolve to "No".
OpenAI’s internal Astra model family, positioned as its next major advancement beyond GPT-5.6 Sol variants, gained attention after an August 1, 2026 disclosure of novel solutions to ten long-standing open problems in mathematics and theoretical computer science, complete with verifiable Lean 4 proofs. This demonstrated strong reasoning on frontier tasks, yet recent evaluations flagged Astra as the first model reaching the “Critical” cybersecurity capability threshold under OpenAI’s Preparedness Framework, enabling autonomous zero-day exploit development. The company paused select training and release steps to implement stronger safeguards, including limited early access via programs like Daybreak Blue. With Astra still fully internal and no confirmed Humanity’s Last Exam submission or public benchmark results amid improving competitor scores from models like Gemini 3 variants, safety-driven timelines and evaluation criteria remain the dominant near-term variables for any debut.