On Google's September 30, 2026 Gemini 4 Argon launch table, Argon scored 55.0% on FrontierSWE v2 versus GPT-6 Astra at 65.5% — a 10.5-point gap, while Argon led DeepSWE v1.1. The split is why a DeepSWE headline is not a universal coding win. Scores are vendor-run unless independently reproduced.