--- format: "aidr-story-markdown/v1" id: "a186da7623d38215b34d262cd3accd7f296ae93b4f5e1386f15bc22dc7a135f8" canonical_url: "https://aidr.today/a186da76?lang=en" title: "OpenAI Internal AI Model's Lowest Setting Beats GPT-6 Astra's Peak on BenchCAD 2.0" lang: "en" requested_lang: "en" available_langs: ["en","vi"] translation_fallback: null fallback_fields: [] published_at: "2026-11-03T23:07:38.000Z" category: "Models" topics: ["openai","reasoning","llm"] source_urls: ["https://huggingnews.com/ai/openai-internal-ai-models-lowest-setting-beats-gpt-6-astras-peak-on-benc-20fcf588","https://marketbrief.now/ai/openai-internal-ai-models-lowest-setting-beats-gpt-6-astras-peak-on-benc-20fcf588"] summary: "An unreleased OpenAI model scored 0.758 on its most basic reasoning level during BenchCAD 2.0 hardware design tests. This score surpassed the 0.594 maximum score recorded by GPT-6 Astra. GPT-6 Astra designed a physical gadget's software, circuit board, and case in less than a week with only two staffers and no direct human edits to the design process." --- # OpenAI Internal AI Model's Lowest Setting Beats GPT\-6 Astra's Peak on BenchCAD 2\.0 > [Open the canonical story]() **Published:** 2026-11-03T23:07:38.000Z **Category:** Models **Topics:** openai, reasoning, llm ## Summary An unreleased OpenAI model scored 0\.758 on its most basic reasoning level during BenchCAD 2\.0 hardware design tests\. This score surpassed the 0\.594 maximum score recorded by GPT\-6 Astra\. GPT\-6 Astra designed a physical gadget's software, circuit board, and case in less than a week with only two staffers and no direct human edits to the design process\. ## Sources - [Story source]() - [Story source]()