Google Unveils Gemini 4 Argon

Argon looks like Google’s clearest attempt yet to compete at the frontier by optimizing for sustained, agentic work rather than short conversational answers.

• –The 1-million-token output limit could enable lengthy software migrations, research workflows, and autonomous debugging loops.
• –Google reports a 77.9% score on DeepSWE v1.1, but independent evaluations will determine whether the gains hold outside Google’s testing setup.
• –Initial access for cyber defenders reflects both the model’s security focus and the risks of releasing stronger autonomous capabilities broadly.
• –The $2/$10 introductory pricing is aggressive, though long reasoning traces and massive outputs could still make task-level costs substantial.
• –Developers should watch API availability, rate limits, tool-use support, and post-introductory pricing before committing production workloads.