Gemini 4 Argon
Points and comments are a snapshot, not live.
Google DeepMind releases Gemini 4 Argon, a new AI model for agents and reasoning.
Google DeepMind launched Gemini 4 Argon, touted as a frontier model for experts in science, math, robotics, and cybersecurity. It features 1M output tokens for uninterrupted reasoning, full-repository code generation, and legacy translation. Argon autonomously turns video into physics-accurate simulations and ties for #1 on CWE-Bench (68%) for vulnerability remediation. Benchmark comparisons show Argon leading in knowledge work (Vals Index 68.9%), agentic coding (DeepSWE v1.1 77.9%), and long-context tasks (GraphWalks 99.7% F1 up to 128k). It is available today for Google AI Pro and Ultra subscribers.
The model powers robotics via Apptronik's Apollo humanoid and cybersecurity via Wiz's 'Scan for Good' initiative. Isomorphic Labs uses Argon to accelerate drug discovery. The release includes safety measures against misuse, prompt injection, and misalignment monitoring.
What commenters are saying
Availability dominated the thread: many commenters reported that Argon was not yet accessible in their region (EU) or subscription tier (Pro), despite the page stating it was available to both Pro and Ultra subscribers. Some later confirmed access appeared after a delay. Several users criticized the announcement page's performance, calling it unusable on mobile Safari and desktop Firefox due to broken scrolling and memory issues.
A secondary thread debated whether AI progress is slowing. One camp argued the delta between releases is shrinking, while another pointed to step-change benchmarks and models solving a Millennium Prize problem as counter-evidence.