source: Google DeepMind: Gemini 4 Argon: our next era of frontier intelligence
level: technical
google deepmind has announced gemini 4 argon, a new frontier model designed for long-horizon, complex professional tasks. it is initially rolling out to a set of trusted cyber defenders through the fairwind program. the model is built to sustain deep reasoning across software engineering, enterprise knowledge work like legal and finance, and cybersecurity defense. google says it is already using argon internally, with thousands of employees reporting strengths in specialized coding, deeper research, and writing quality.
argon introduces an industry-leading 1 million token output limit, up from 64k tokens, allowing it to generate hundreds of thousands of tokens in a single trajectory. on the deepswe v1.1 benchmark for real-world software engineering, it scores 77.9 percent. it ranks first on automationbench with 51.3 percent and on lvbench for long video understanding with 91.7 percent. on cwe-bench v1 for vulnerability remediation, it ties for first at 68 percent. pricing starts at $2 per million input tokens and $10 per million output tokens.
google is taking a phased approach to release, engaging with the u.s. government's voluntary process for pre-release model access. the model is being released without cyber guardrails to trusted defenders, allowing full use for finding and patching vulnerabilities. wiz is already using argon through its scan for good initiative, where it uncovered a critical vulnerability in healthcare software that previous models missed. google is also strengthening safeguards against misuse, prompt injection, and misalignment before broader availability.
why it matters: argon's 1m token output and strong benchmark scores could change how ai handles long, multi-step coding and security tasks, but its limited release means most users cannot access it yet.
source: Google DeepMind: Gemini 4 Argon: our next era of frontier intelligence