Gemini 4 Argon Challenges Frontier Coding Leadership

Google’s highly restricted September 30 release claims strong coding and agentic performance, including DeepSWE leadership, autonomous vulnerability discovery and patching, a 1-million-token output ceiling, and internal Rust migration results. Independent validation is sparse, and weaker results on FrontierSWE, Terminal-Bench, and some terminal workloads reinforce that benchmark and harness choice—not headline leadership alone—should guide model selection.