← Back to feed

Gemini 4 Argon moves from restricted to general availability on October 13. Google launched the model on September 30 through the Fairwind Program, limiting access to cybersecurity defenders. Full global availability now covers developers and enterprises via the Gemini API and AI Studio, Google AI Ultra subscribers through Google Antigravity, and both Ultra and Pro subscribers in the Gemini app and AI Mode in Search.

Robotics: 87% on ERQA

The standout new benchmark number is 87% on the Embodied Reasoning Quality Assessment (ERQA), which tests a model’s ability to understand and solve physical, spatial problems in 3D environments using pure reasoning without robotics hardware. Argon leads all evaluated frontier models on this benchmark.

The practical application: Argon can take a raw video of a human performing a physical task, extract physical coordinates, and generate simulation code without the months of manual coding that robotic workflow pipelines traditionally require. Google positions this as a step toward physical AGI — systems that can reason about the world in three dimensions, not just language and images.

The robotics capability combines three underlying strengths: advanced visual-semantic reasoning (reading a scene accurately), deep physical understanding (modeling how objects move and interact), and autonomous code generation (turning that understanding into runnable simulation or control code).

1M Unified Token Limit

Argon ships with a 1 million token unified input-output limit via the Gemini API — industry-leading on the output side. Earlier frontier models cap output at 64K tokens; Argon’s limit is roughly 15 times larger. Google has added a continuation protocol for developers and enterprises that need multi-stage outputs beyond a single context window.

This matters most for the professional workloads Argon targets: generating comprehensive legal documents, processing long financial reports, writing extensive codebases in a single inference call, or doing multi-step scientific analysis without mid-task interruption.

Pricing: Intro Discount Expires November 30

Introductory pricing through November 30, 2026: $2 per million input tokens, $10 per million output tokens. Cached input tokens are priced at 95% off the input rate during the intro period.

After the introductory period, pricing moves to $4 per million input tokens and $20 per million output tokens — the figures that have appeared in benchmark cost analyses since the Fairwind launch. Developers running agentic and multi-step workloads have roughly seven weeks to benchmark at the discounted rate.

Where It Leads

Argon holds the top position on the Vals Index — a benchmark weighted by GDP share across finance, coding, legal, and tax — ahead of 43 evaluated models. It also leads Arena.ai’s Text leaderboard at 1,525 ELO and posts the lowest hallucination rate of any high-intelligence frontier model at 15% on Artificial Analysis’s AA-Omniscience factual benchmark.

The full-access launch extends these capabilities to consumer and developer use cases that the Fairwind restriction excluded. The priority order on the access timeline — cybersecurity first, then general release — reflects the design logic: a model optimized for precision over recall, with a hallucination rate calibrated for professional contexts where wrong answers are expensive.