Holy shit, if all the benchmarks are true, then GPT-6 Astra is a pretty big deal, and probably bigger than we all realize. 98.6% on ARC-AGI-3 is a massive jump from 7.8% six months ago. 100% on ExploitBench and 97.6% on FrontierMath Tier 4 are genuinely frontier-pushing. Astra had a much higher arbitrary-code-execution rate than GPT-5.6 Sol while using significantly fewer output tokens. And then something pretty wild happened: Astra discovered and used two zero-day vulnerabilities as part…