Gemini 4 Argon launches with up to 1M output tokens per trajectory
Google introduced Gemini 4 Argon, offering up to 1M output tokens per trajectory for software engineering, legal, finance and security work, scoring 77.9% on DeepSWE v1.1 with API pricing from $2/$10.
Google just introduced Gemini 4 Argon. This looks like the big frontier model of the Gemini 4 generation, built for long, difficult workflows rather than quick everyday prompts. A few things immediately stand out: → Up to 1 MILLION output tokens in a single trajectory → Strong focus on real software engineering, legal, finance, research, and cybersecurity → Google says it reaches 77.9% on DeepSWE v1.1 and currently leads several professional workflow benchmarks → Argon is already being used internally at Google for large code migrations, optimization work, research, and autonomous engineering tasks → It can autonomously find, validate, and patch software vulnerabilities → Introductory API pricing starts at $2/M input and $10/M output, with cached input heavily discounted → It will first roll out to trusted testers, cyber defenders, paid API users, and Google AI Ultra subscribers before expanding more broadly This is probably the most interesting Gemini announcement in a while. Now I want to see how it actually performs outside Google’s own benchmarks.