Google’s Gemini 4 Argon Closes the Pricing Triangle. The Benchmark Lead Is the Real Story.

Google DeepMind chose the last day of September to do something that makes the competitive landscape of frontier AI look almost mathematical. On September 30, SVP Koray Kavukcuoglu announced Gemini 4 Argon at introductory pricing of $2 per million input tokens and $10 per million output tokens, with cached input at 95% off. Twenty-four hours earlier, OpenAI had launched GPT-6.1 Sol at exactly the same rates.

The match is not a coincidence. It is the commodity floor arriving in the frontier model tier. OpenAI halved its own pricing from GPT-6 Astra to get here. Google, entering the race for the first time with a model capable of leading on benchmarks, chose to meet that price rather than undercut or exceed it. The introductory rates will eventually double to $4/$20 – but for now, the signal is clear: the two largest US labs have converged on the same economics for near-frontier intelligence.

Anthropic's S-1 prospectus, filed two days earlier with $518 billion in compute commitments, makes the contrast sharper. Anthropic has not launched a $2/$10 competitor. Its strategy remains enterprise-led, anchored by Claude Code and quarterly revenue of $11.6 billion that now surpasses OpenAI's. The pricing triangle – OpenAI at $2/$10, Google at $2/$10, Anthropic enterprise-only – represents a genuine bifurcation in business models, not merely a difference in marketing.

Google is positioning Argon as the new frontier leader, and the number it is leading with is DeepSWE v1.1: 77.9%. That score, if it holds under independent scrutiny, edges out Claude Opus 5.5 at 74.2% and GPT-6 Astra at 74.1%. DeepSWE measures real-world long-horizon software engineering tasks – not synthetic benchmarks, but the kind of multi-step coding work that enterprise buyers actually need.

The caveat is significant. These are vendor-reported figures. Google has not published independent verification, and the industry's track record on self-reported benchmarks is uneven. What is verifiable is the model's technical specifications: a 2 million token context window and a 1 million output token limit, up from 64,000 in prior Gemini models. That output ceiling is industry-leading and designed for the kind of sustained, complex reasoning that enterprise workflows demand.

Google reports that internal use of Argon has already produced a 40% improvement in quantum algorithmic optimization, freed over 300 TiB of memory across its data centers, and is driving large-scale codebase migrations to Rust – including an 800,000-line rewrite of the Fuchsia Zircon kernel. On the Vals Index , which measures economic impact across finance, coding, legal, and tax work weighted by GDP contribution, Argon ranks first. On CWE-bench v1 , which evaluates vulnerability remediation, it ties for first at 68%.